# Consumer Search, Price Dispersion, and Asymmetric...

### Transcript of Consumer Search, Price Dispersion, and Asymmetric...

University of California

Los Angeles

Consumer Search, Price Dispersion, and

Asymmetric Pricing

A dissertation submitted in partial satisfaction

of the requirements for the degree

Doctor of Philosophy in Economics

by

Mariano Emilio Tappata

2006

c© Copyright by

Mariano Emilio Tappata

2006

The dissertation of Mariano Emilio Tappata is approved.

Daniel Ackerberg

Bryan Ellickson

Sushil Bikhchandani

David K. Levine, Committee Co-chair

Hugo Hopenhayn, Committee Co-chair

University of California, Los Angeles

2006

ii

To my wife Florencia, my sons Felipe and Benjamın

and especially to my parents

iii

Table of Contents

1 Rockets and Feathers. Understanding Asymmetric Pricing. . . 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 The model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.3 Dynamics and asymmetric pricing . . . . . . . . . . . . . . . . . . 18

1.4 More sellers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

1.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

Appendices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

A - Proofs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

B - Multiunit Demands . . . . . . . . . . . . . . . . . . . . . . . . 38

C - Sequential Search . . . . . . . . . . . . . . . . . . . . . . . . . 48

2 Price Dispersion and consumer search.

The Retail Gasoline Markets. . . . . . . . . . . . . . . . . . . . . . . . 60

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

2.2 Consumer Search and Price Dispersion . . . . . . . . . . . . . . . 63

2.3 The gasoline market . . . . . . . . . . . . . . . . . . . . . . . . . 67

2.4 Price dispersion analysis . . . . . . . . . . . . . . . . . . . . . . . 69

2.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

iv

List of Figures

1.1 Equilibrium with uniformly distributed search costs . . . . . . . . 14

1.2 Market equilibrium with shoppers and homogeneous nonshoppers 16

1.3 Equilibrium price distribution and number of stores (c = 0, µ = 0.2) 27

1.4 Maximum expected gains from search and the number of firms . . 29

1.5 Cost pass-through . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

1.6 Consumer demand and firms’ NBR . . . . . . . . . . . . . . . . . 40

1.7 Gains from search and price dispersion with linear demands. . . . 43

1.8 Higher production costs. Constant elasticity demands. . . . . . . 45

1.9 Numerical simulations for θ > 1 . . . . . . . . . . . . . . . . . . . 46

1.10 Updated prior and unique reservation price . . . . . . . . . . . . . 55

2.1 Rank reversals in prices . . . . . . . . . . . . . . . . . . . . . . . 72

2.2 Cumulative rank reversals and distance . . . . . . . . . . . . . . . 73

2.3 Price level and dispersion . . . . . . . . . . . . . . . . . . . . . . . 76

2.4 Percentage of articles published in the AER, JPE, and Economet-

rica on Information, Search or Price Dispersion . . . . . . . . . . 78

2.5 Industry structure . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

v

List of Tables

1.1 Market equilibrium with shoppers and homogeneous nonshoppers 18

1.2 Expected gains from search . . . . . . . . . . . . . . . . . . . . . 29

1.3 Maximum E[p− pmin|c, n] and n . . . . . . . . . . . . . . . . . . . 29

2.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . 70

2.2 Kolmogorov-Smirnov equality of distributions test . . . . . . . . . 75

2.3 Models of search . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

vi

Acknowledgments

I would like to thank many people for all their support during the dissertation

process. Starting with my advisors, I have only gratitude to my co-chairs David

K. Levine and Hugo Hopenhayn, who -in different ways- helped me enormously

in each step of the way. David kindly accepted to be the chair of my dissertation

even though my original projects were only tangential to his research interests.

He was nevertheless, always very supportive and provided invaluable guidance

every time I was not sure how to approach certain economic problems. I am

also grateful for his patience and dedication with grant aplicactions and during

the job market process. Hugo’s arrival at UCLA changed the direction and -

most importantly- the speed of my work. Every talk we had was particularly

illuminating, and I think I learned more of Industrial Organization during these

talks than ever before. He was constantly pushing me to think about important

issues and redirected my work when I would try to solve obvious (to him!) and

therefore minor problems. I hope some day to reach only a fraction of his clear

and rigorous way of approaching the study of economics. His kindness expanded

beyond the academics, and Hugo -as well as Estela- were very generous with me

and my family.

I would also like to thank the other members of my committee for their help

and encouragement. Dan Ackerberg provided invaluable insights on my empirical

projects and taught me the most interesting course during my PhD. He has the

gift of identifying the roots of any result and was kind enough to share it with me.

Although there is not a lot of empirical work on this dissertation, I only blame

it on Dan’s stay in Arizona. However, I am sure his influence on my research

will become more explicit in my future work. In retrospect, I regret not having

vii

interacted more with both Bryan Ellickson and Sushil Bikhchandani. The few

times that I asked for their help, they were incredibly open and welcoming, as

well as insightful. I also would like to thank all my UCLA professors from whom

I benefited being their student and/or talking after my presentations. In particu-

lar Hongbin Cai, Harold Demsetz, Moshe Buchinsky, John Riley, Bill Zame, and

Jean-Laurent Rosenthal.

I have made great friendships at UCLA and learned a lot from each one of

them. Thanks to Mauro, Mati, Guille, mono, Roberto, Juan Manuel, David

B.,Juan , Rolf, David R., Christine, and Miguel. I also want to thank Santiago

Urbiztondo, Walter Cont, Jose Wynne and Alito Harberger. Without their help,

I would have never started my graduate studies at UCLA.

These past five years have been a wonderful and fulfilling experience. They

would not have been possible without the relentless support from my family.

Thanks to uncle Henry and our Californian family, things were so much easier

with their support. Thanks to my wife Florencia for her patience and under-

standing, she gave me strength during the most difficult times and helped me

read and revise my manuscripts innumerable times. Last, but most importantly,

I am eternally grateful to my parents, Heber and Anahı, who taught me the

power of sacrifice and shared their passion for economics.

viii

Vita

March, 16, 1973 Born, Bahıa Blanca, Argentina

1998 Licenciado (Economics)

Universidad Nacional de La Plata

La Plata, Argentina

1996–1999 Advisor of the Secretary of Fiscal Policy

Ministry of Finance, Province of Buenos Aires

La Plata, Argentina

1998–2000 Lecturer

Universidad Nacional de La Plata

La Plata, Argentina

1999–2000 M.A. (Finance)

Universidad Torcuato Di Tella

Buenos Aires, Argentina

2000 Best Teaching Assistant Award

Master in Finance

Business School, Universidad Torcuato Di Tella

Buenos Aires, Argentina

1999–2001 Junior Researcher

Center for Financial Research

Universidad Torcuato Di Tella, Argentina

2001–2002 Bradley Foundation Fellowship

ix

2003 M.A. (Economics)

UCLA

Los Angeles, California

2003 Best Teaching Assistnat Award

Department of Economics

UCLA

Los Angeles, California

2003 C. Phil. (Economics)

UCLA

Los Angeles, California

2004–2005 University of California Energy Institute Grant

2002–2006 Teaching Assistant/Associate/Fellow

Department of Economics

UCLA

Los Angeles, California

2006 Robert Ettinger Prize (best paper by a UCLA Economics De-

partment graduate student).

2006–present Assistant Professor

Strategy and Business Economics Division

Sauder School of Business

University of British Columbia

Vancouver, Canada

x

Publications and Presentations

Lodola, Agustin and Mariano Tappata 1997. “ Local Government’s Debt. The

case of the Municipalities in Buenos Aires.”Anales Asociacion Argentina de Fi-

nanzas Publicas, XXX Annual Meetings, Cordoba, Argentina.

Tappata, Mariano, Gustavo Jaconiuk and Eduardo, Levy Yeyati 2000. “The

Use of Derivatives by Non-Financial Argentinean Firms,”CIF Working Paper #7

Universidad Torcuato Di Tella, Argentina.

“Rockets and Feathers. Understanding Asymmetric Pricing.” Paper presented at

the Midwest Economic Theory Meeting, U. of Kansas, Lawrence, October 2005;

the International Industrial Organization Conference, Northeastern University,

Boston, April 2006; and the LACEA-LAMES meetings, ITAM, Distrito Federal

(Mexico), November 2006.

xi

Abstract of the Dissertation

Consumer Search, Price Dispersion, and

Asymmetric Pricing

by

Mariano Emilio Tappata

Doctor of Philosophy in Economics

University of California, Los Angeles, 2006

Professor Hugo Hopenhayn, Co-chair

Professor David K. Levine, Co-chair

This dissertation consists of a study of the consequences of consumers’ imperfect

information on market clearing prices. The traditional paradigm in economics as-

sumes consumers have perfect information about the prices in the market. When

this assumption is replaced by the more realistic one of costly information acqui-

sition (consumer search) the predictions from the perfectly competitive market

change radically. Price dispersion emerges even when firms are identical and sell

homogeneous products. Moreover, profits or information rents are captured by

the sellers in the long run.

In Chapter I, I explore the theoretical implications of consumer search on price

dynamics. Previous empirical work established that in most markets “prices rise

like rockets but fall like feathers.” I show that a model with competitive firms and

rational partially-informed consumers can generate such asymmetric response to

costs by firms. In contrast to public opinion and past work, collusion is not

necessary to explain such stylized fact.

In Chapter II, I analyze the price dispersion observed in the Californian retail

xii

gasoline markets. The retail gasoline market presents a unique opportunity to

identify the sources of price dispersion. The price differences between gas stations

located in a single corner can only be related to product characteristics, while the

price spreads between stations that are further appart can also be generated by

costly consumer search. Using a rich and unique dataset on retail prices, I show

that consumers’ imperfect information is important in this market.

xiii

CHAPTER 1

Rockets and Feathers. Understanding

Asymmetric Pricing.

1.1 Introduction

Output prices do not react symmetrically to changes in input prices. According to

Peltzman’s comprehensive study of 165 producer goods and 77 consumer goods,

“In two out of three markets, output prices rise faster than they fall”(Peltzman,

2000; p. 480). This pattern is also known as rockets and feathers and has

sometimes been used interchangeably with the term asymmetric pricing.1 De-

spite the abundance of empirical work confirming this stylized fact, there has

not been much progress in terms of theoretical explanations for this widespread

phenomenon.

The first thing that comes to mind when talking about rockets and feathers

is collusion. A classical example is gasoline retailing, a market operated by a

handful of players with output and input prices easily observable by everyone.

Asymmetric gas price adjustments are usually associated with collusive behavior

by both government and the media.2,3 However, Peltzman finds that the rockets

1To the best of my knowledge, Bacon (1991) was the first to use the term rockets and feathersto describe the pattern of retail gasoline prices in the U.K.

2See Karrenbrock (1991; p. 20) for media and government representative quotations aboutgasoline price gouging.

3This perception, together with a lack of input substitution possibilities in gasoline produc-

1

and feathers pattern is equally likely to be found in both concentrated and atom-

istic markets. In this paper, I develop a consumer-search model that explains

how an asymmetric response of prices to costs can arise in competitive markets.

According to traditional economic theory, homogeneous firms that compete

on prices earn zero profit, and cost shocks are completely transferred to final

prices.4 The nature of this equilibrium changes drastically if consumers are im-

perfectly informed of market prices and a fraction of them has positive search

costs. Competitive firms now profit from informational rents, and equilibrium

is characterized by price dispersion instead of a single price. Still, for any given

level of production costs, firms’ optimal price margin is the same regardless of

whether their cost shock was positive or negative. In order to obtain asymmetric

pricing, the demand function faced by the firms must be sensitive to previous

cost realizations. This is indeed what happens when consumers don’t observe

firms’ current production cost.

I introduce uncertainty over production costs in a nonsequential search model

similar to Varian’s model of sales (Varian, 1980). Given consumers’ search in-

tensity, firms maximize profit by choosing prices that are less dispersed under

high than low production costs, since their scope to set prices -measured by the

gap between marginal cost and the monopoly price- decreases. Rational con-

sumers anticipate this and therefore search less when they expect costs to be

tion, influenced the focus of most empirical work (Bacon, 1991; Karrenbrock, 1991; Borenstein,Cameron, and Gilbert, 1997; Lewis, 2003; Deltas, 2004; and Verlinda, 2005 among others).Empirical research investigating asymmetric pricing in other markets includes Neumark andSharpe (1992) and Hannan and Berger (1991) in the banking sector; and Boyd and Brorsen(1998), and Goodwin and Holt (1999) in the food industry.

4Although firms with market power and costless consumer search don’t transfer all of theircost shocks to consumers, they still price symmetrically in that the price they optimally chargedepends only on current cost realizations, not on previous costs. Therefore, the rate of change inprices is always the same (as a function of costs) regardless of previous prices, which eliminatesthe possibility of rockets and feathers

2

high. Intuitively, when input cost shocks are not independent over time, con-

sumers’ expectations differ depending on whether cost was high or low in the

previous period. This translates into different demand elasticities faced by firms

when cost falls or rises and therefore, prices react asymmetrically to cost shocks

as the firms’ pass-through increases with the level of competition in the market.5

The rockets and feathers pattern emerges under persistent cost realizations.

Suppose that the current marginal cost is high. Consumers expect it will remain

high, so they expect little price dispersion and search very little. If in fact the

unexpected occurs and marginal cost drops, firms have little incentive to lower

their prices because consumers aren’t searching very much. On the other hand,

if marginal cost is currently low, it is likely to stay low, so next period price

dispersion is expected to be high, consumers search intensifies, and the response

by firms to a positive cost shock is to raise prices significantly.

This paper links asymmetric pricing in competitive markets with costly con-

sumer search. A general characteristic of consumer search models is price disper-

sion. However, this pattern is also consistent with models of product differentia-

tion in the market. Whether the widespread price dispersion observed in many

markets is attributable to consumer search, product differentiation, or both is an

empirical question. In Chapter 2, I show that the retail gasoline market (a mar-

ket where evidence of rockets and feathers has been found many times) exhibits

price dispersion consistent with both, product differentiation as well as costly

consumer search.

The contribution of this paper is in formalizing a model with rational agents

that isolates the crucial features needed for asymmetric pricing to emerge in

5The extension of the model to the case of multiunit demands is analyzed in Appendix B.

3

competitive markets. The most related work is represented by Lewis (2003). He

develops a reference-price search model with homogeneous firms and consumers

that form adaptive expectations about the current price distribution. Consumers

search sequentially and their search strategies are optimal with respect to past

reference prices, although not necessarily to actual prices. Firms then use this

myopic behavior to their advantage and set prices to minimize search by con-

sumers. If costs drop below past price, firms need to only decrease their prices a

little to avoid search, while if cost increases above past prices, there is no option

but to set prices at least as high as the new cost, which in equilibrium generates

consumer search.6 In this paper consumers use all available information to them.

In that sense, the approach is similar to Benabou and Gertner (1993). They

study the effect of inflation’s uncertainty on efficiency in a market composed

by consumers that search sequentially and heterogeneous firms with production

costs composed of both an idiosyncratic (real) and a common (inflation) shock.

Consumers behave rationally by updating their priors about the common shock

from observed prices. Under some parameters, more inflation uncertainty leads

to more search and thus generate inefficiencies.7,8

This model shares the assumption that consumers are imperfectly informed

6In this case, after visiting n−1 stores and observing n−1 identical prices, consumers wouldstill choose to pay the search cost and sample from the nth store since they believe that theprices in the market are normally distributed with a mean lower than the observed price.

7Borenstein et al. (1997) suggest a reinterpretation of this model to account for asymmetricpricing. If changes in the (common) production cost imply higher volatility, less search isrelated to higher and lower costs. Firms can charge a higher mark-up due to lower search andthe cost pass-through is bigger (smaller) if cost is increasing (decreasing).

8Other work on asymmetric pricing is Borenstein et al. (1997) and Eckert (2002). Theformer suggest a model of tacit collusion with imperfect monitoring (as in Tirole, 1988; p.264). With multiple equilibria, firms collude using the past-period price as a focal point.Decreases in production cost facilitate coordination on previous price, while if cost increases itis likely that past price is unprofitable, collusion breaks down and a higher price emerges asa new equilibrium. On the other hand, Eckert uses a model of Edgeworth cycles to explaingasoline price movements that are independent of cost shocks. This pattern has been observedin some Canadian cities.

4

with Benabou et al. (1993) and Lewis (2003). In contrast to their work, I

assume homogeneous firms (as Lewis), agents that form rational expectations

(as Benabou et al.), and consumers searching nonsequentially. That is, each

consumer decides -before observing any prices- between becoming informed about

all market prices (and buying from the store with the lowest price) or remaining

uninformed, in which case she buys costlessly from a random store. If a consumer

were to search sequentially, after visiting a store she would decide whether to

sample for another price or shop at the lowest price observed at that moment.9

The early literature on consumer-search models focused on nonsequential

search protocols (Salop and Stiglitz, 1977; Braverman, 1980; and Varian, 1980),

while more recently sequential search models have dominated the literature (Stahl,

1989 and 1996; and Benabou and Gertner, 1993). Both sequential and nonse-

quential search protocols can be optimal depending on the context of the decision

problem (Morgan and Manning, 1985).10 Nonsequential search tends to dominate

when price quotes are not obtained instantaneously (insurance quotes, repair es-

timates, etc.), the opportunity cost of time is relatively high, and when there are

economies of scale in the size of the price sample (online shopping). When price

quotes are obtained easily and there are no economies of scale, sequential search

tends to dominate nonsequential search protocols, since it allows consumers to

stop searching as soon as they find a good bargain.11

The rest of the chapter is organized as follows. In the next section I describe

9This is the case of sequential search with perfect recall. In the case of no recall, if theconsumer stops searching, she must shop at the last observed price.

10Other search protocols have been used as well. Dana (1994) uses a mixture of sequentialand nonsequential search. After a consumer observes a first price she needs to decide if shewants to pay to know the rest of the prices in the market. Burdett and Judd (1983) assume aflexible sample-size nonsequential search protocol.

11Appendix C contains an extension of the model for the case of sequential search. I showthat a necessary condition for asymmetric pricing is the existence of heterogeneous search costs.

5

the model and the static duopoly equilibrium. Next, the dynamic setting is

introduced together with the rockets and feathers result. In section IV, I extend

the result to markets with more than two firms. Section V concludes.

1.2 The model

In this section I lay out a static duopoly model where firms compete choosing

prices and consumers decide whether to search or not based on some prior over

firms’ production costs. The model is an extension of Varian’s model of sales

(Varian, 1980) where I endogenize consumers’ search decisions and incorporate

uncertainty over production costs. The two main results of this section are the

following: First, the market equilibrium involves price dispersion and a fraction

of consumers choosing to become informed (Proposition 2). Second, the search

intensity in the market decreases with the expected production cost (Lemma 2).12

This static model serves as the stage game in a dynamic model that I introduce

in the next section.

Consider two firms with the same marginal and average production cost selling

a homogeneous good. At the beginning of the period, Nature draws the cost

for the industry, firms observe the cost realization and compete through prices.

There is a continuum of consumers of measure one who only know the probability

distribution of the marginal cost. To simplify the analysis, assume they each have

a unit demand with a choke price υ, and can obtain information about market

prices through nonsequential search.13 They decide -before observing any prices-

12Dana (1994) analyzes the effects of consumer learning in a static model where with incom-plete information about the firms’ cost of production. For the duopoly case the search protocolused there by consumers is equivalent to sequential search (see footnote 10).

13Consumers with unit demand is a simplifying assumption and is not critical for the rocketsand feathers result. Intermediate results change for some demand functions and I discuss that

6

between becoming informed and buying from the store with the lowest price,

or shopping at a randomly selected store. Nonsequential search protocols are

especially appealing to consumers when there are economies of scale in price

sampling. Products that are advertised in weekly newspapers are a classical

example of such advantages. More recent examples include specialized websites

that aggregate and compare all the relevant information across online stores, and

that save consumers the trouble of a sequential search.14

The cost of becoming informed is the search cost. Assume that a portion

λ ∈ (0, 1) of the consumers has zero or negative search cost and I refer to them

as shoppers. Shoppers can be interpreted as consumers who enjoy searching for

prices or who have obtained price information unintentionally through advertising

or while shopping for other goods. The remaining (1− λ) consumers have positive

search costs that are drawn from a continuous and differentiable cdf g (si), with

si ∈ S = [0, s] and s > υ.

Given the nature of the search protocol, consumers and firms decide their

actions simultaneously. The search/no search decision by consumers will be af-

fected by the expected price dispersion in the market and their search costs. So

based on their priors about the marginal cost realization, consumers form ratio-

nal expectations on firms’ pricing strategies to forecast price dispersion. At the

same time, firms set their prices anticipating the search intensity in the market.

More formally, firms and consumers play a simultaneous-move Bayesian game

with N =NF ∪ND

players, where j ∈ NF = I, II denotes a firm and

in the text. Appendix B extends the results to two other type of demands (linear and constantelasticity).

14Another example where nonsequential search is optimal is that of daily commuters decidingwhere to buy gasoline (see more in Chapter 2).

7

i ∈ ND = [0, 1] a consumer. Producers can be of either type cL or cH , where the

probability of having high cost (type cH) is α. Consumers’ search costs (or their

types) si ∈ S are public knowledge. Firms choose prices pj in the interval P =

[cL, υ] and consumers choose actions ai ∈ A = 0, 1 = don’t search, search .15

Letting µ =1∫0

aidi represent the number of informed consumers, the profit of a

firm j that charges a price pj and has production cost c is given by:

πj (pj, p−j, a, c) = (pj − c)

1 + µ

2Ipj<p−j +

1

2Ipj=p−j +

1− µ

2Ipj>p−j

(1.1)

where p−j represents the price charged by firm j’s competitor, and I is an indicator

function. Meanwhile, the conditional utility of a consumer i with search cost si

is:

ui (ai, a−i, p) = υ − ai (Min [p] + si)− (1− ai)1

2

∑j

pj (1.2)

Firm j′s strategy profile is represented by all possible price distributions given

a cost realization: fj (·, c) = fj (pj, c)pj∈P with fj (pj, c) ≥ 0 for all pj ∈ P and∫Pfj (p, c) dp = 1. Consumers on the other hand have strategy profiles qi (·, si) ∈

∆ (A) that include the possibility of randomizing between search and no search.

The interaction between consumers and firms can be summarized by the

proportion of informed consumers µ. Any strategy profile for the consumers

σD = qi (·, si)i∈ND implies a value of µ ∈ [λ, 1].16 Define a Nash Best Re-

sponse NBR (µ, c) as a symmetric Nash Equilibrium strategy of the game Γ =

[NJ , P, πj∈NJ] where πj is defined in (1.1). That is, a NBR consists on the equi-

librium price strategies in the duopoly game that are a best response to a given

15I ignore the decision between buying or not for the consumer by setting υ as the upperbound for pj . This simplifies notation and does not affect any result.

16This is consistent with the definition of shoppers given above. If shoppers are thought ofas consumers with zero search cost, I break any potential indifference in (1.2) by assuming theyalways search.

8

search intensity by consumers. A Symmetric Bayesian Nash Equilibrium (SBNE)

or market equilibrium is composed of consumers’ beliefs about the marginal cost,

α and a strategy profile σ =(σD, σF

)such that i) σD is a best response to

σF = (f (p, c, µ))p∈P and ii) σF is a NBR(µ(σD), c). In words, a market equi-

librium is characterized by consumers that search optimally given the pricing

strategies of the firms, and firms that set prices optimally given the number of

consumers that become informed.

Start analyzing the supply side of the model by obtaining the firms NBR. A

given number of informed consumers µ can be related to the expected elasticity

of demand faced by each firm. This is clear when we examine the extreme cases

of µ = 0 and µ = 1. The former corresponds to two separate monopolies. Each

firm faces a completely inelastic demand and maximizes profits by extracting

all the consumer surplus (p = υ). On the other hand, when all consumers are

informed about the market prices (µ = 1) , firms face perfectly elastic demands

which leave them no option but to price at marginal cost. In the rest of the

cases (0 < µ < 1), each firm faces an expected downward slopping demand. It is

easy to verify that there is no single price equilibrium (SPE) since a store would

capture the informed consumers µ by slightly undercutting its competitor.17

The assumptions made on consumers’ search costs eliminate the possibility of

monopoly or perfect competition outcomes. First, a lower bound on the number

of informed consumers is given by the number of shoppers in the market (µ ≥ λ).

On the other hand, as will be seen below, the existence of consumers with high

search cost (s > υ) implies that there is always a mass of uninformed consumers in

17Note that SPE and pure strategy equilibrium are equivalent since NBR is defined to be asymmetric NE.

9

equilibrium.18 Therefore, given µ, a firm with cost c that sets a price p can either

fail or succeed in capturing the informed consumers. Its profits are respectively:

πf (p, c) =(1− µ)

2(p− c) (1.3)

πs (p, c) =(1 + µ)

2(p− c) (1.4)

By charging the highest possible price, a firm can always guarantee itself a

positive profit equal to the surplus of its captive consumers:

π(υ, c) =(1− µ)

2(υ − c) (1.5)

This, places a lower bound on the prices considered by any firm. Even if a firm

captured all the informed consumers, charging a price below p∗ generates less

profits than if it charged the monopoly price:19

p∗ = πs −1

(π(υ, c)) = c+(1− µ)

(1 + µ)(υ − c) (1.6)

Thus, a NBR consists of strategies over [p∗, υ] .

By the same argument used to ruled out any single price equilibrium, all

mixing strategies that involve a positive mass over any price can be ignored.

Denote the cumulative distribution implied by a particular strategy profile σF

with F (·, c, µ). A firm is indifferent between charging the monopoly price and a

price that generates a similar expected profit:

πs(p, c)(1− F (·)) + πf (p, c)F (·) = π(υ, c) (1.7)

18The perfect competition outcome will actually not arise as long as there is a proportion ofconsumers with positive search cost (not necessarily > υ). This is because when µ = 1 firms setprices equal to the production cost. Then, since it would never pay to consumers with positivesearch cost to search (no price dispersion), µ = 1 is a contradiction.

19Note that by definition p∗ cannot be a SPE.

10

High prices increase mark-ups per unit sold but decrease the expected market

share by reducing the likelihood of being the cheapest firm in the market. The

surplus-appropiation and business-stealing effects characterize the trade-off faced

by firms, which induces price dispersion or the existence of sales (Varian, 1981).

Proposition 1 There is a unique Nash Best Response σF . Given µ and c, the

cumulative distribution of market prices is

F (p, c, µ) =p∫p∗f (x, c, µ) dx = 1−

((1− µ)(υ − p)

2µ(p− c)

)(1.8)

for all p ∈[p∗ = c+ (1−µ)

(1+µ)(υ − c), υ

]Proof: See Appendix A.

The share of informed consumers affects the pricing strategies of the firms in

two ways. First, as µ increases, there is a smaller captive market for each firm

and the profit made by charging the monopoly price decreases. This increases the

equilibrium range of prices over which firms are willing to randomize in order to

attract the informed consumers (equation 1.6). At the same time, a larger pro-

portion of informed consumers makes the business-stealing effect more attractive,

hence relatively more weight is placed on low prices. This can be seen in (1.8) as

F (·.µ′) first-order stochastically dominates F (·.µ) when µ′ > µ.

On the demand side, consumers decide between becoming informed about the

market prices (at a cost si) or buying from a random store. The market demand

is composed of consumers whose individual choices ai do not influence the search

intensity in the market. Given the firms’ NBR σF , the expected benefit for each

consumer of being informed is measured by the difference between the expected

11

price and the expected minimum price in the market (price dispersion):

E [p− pmin|µ] = Ec

v∫p∗

p [1− 2 [1− F (p, c, µ)]] dF (·, c, µ)

=

= (υ − E [c])(1− µ)

2µ2

[log

[1 + µ

1− µ

]− 2µ

](1.9)

where the last equation is obtained using (1.8) and integrating by parts.

Expected market price dispersion is what drives consumer to search. At the

same time, price dispersion depends on the amount of informed consumers. Start-

ing from a monopoly situation with µ = 0 and no price dispersion (p = υ), as

µ increases, firms start choosing prices over a wider range of prices and plac-

ing relatively more likelihood on low prices. This has the effect that both, the

expected price and the expected minimum price decrease. But they do it at dif-

ferent rates and there exists an amount of informed consumers µ at which the

consumers’ gains from search are maximized.20 For µ > µ, adding informed con-

sumers reduces the spread between the expected price and minimum price since

the firms increase the probability of choosing low prices while keeping the domain

in (1.8) relatively fixed. The following lemma characterizes the price dispersion

as a function of the search intensity (equation 1.9).

Lemma 1 The consumers’ expected gains from search is a strictly concave func-

tion of the number of informed consumers. Furthermore, it has a maximum at

µ ∈ (1/2, 1) .

Proof: See Appendix A.

20E [p] decreases at a decreasing rate for any µ while E [pmin] does it at an increasing ratefor µ < 0.78341 and a decreasing rate for bigger µ’s.

12

Consumers compare the benefits from becoming informed to their search costs.

Thus, shoppers always search for low prices while consumers with search cost

higher than υ never search.21 That also implies that there are at least λ in-

formed and (1− g (υ)) (1− λ) uninformed consumers in a market equilibrium.

For the remaining consumers, the optimal search strategies are qi (si < s) = 1

and qi (si > s) = 0 where s is the search cost of the indifferent consumer:

E[p− pmin|µ = λ+ (1− λ) g (s)]]− s = 0 (1.10)

A market equilibrium when consumers have uniformly distributed search costs

is shown in Figure 1.1. The proportion of informed consumers is measured on

the horizontal axis, while the search costs and gains from search are on the

vertical axis. The dashed and solid concave curve represents the gains from

search to consumers. Each consumer compares her search cost with the gains

from search given the total amount of informed consumers. The straight line

with positive slope represents the search cost of the marginal consumer that

decides to search. The unique equilibrium is represented by the intersection of

the two curves. Consumers with search cost lower than s search and those with

higher cost choose to remain uninformed.

A unique equilibrium is obtained under any search cost distribution as long

as there is a large number of shoppers (λ > µ). When this is not the case,

there could be more than one solution to (1.10) depending on the slope of the

curves representing the search cost of the marginal consumer and the gains from

search. The next proposition states the conditions required for a unique market

equilibrium.

21See footnotes 18 and 16.

13

Figure 1.1: Equilibrium with uniformly distributed search costs

Proposition 2 There is a unique market equilibrium if:

a) λ > µ, or

b) 0 < λ < µ and ∂g−1

∂µ> ∂E[p−pmin]

∂µover µ ∈ [λ, µ] .

Proof: See Appendix A.

The market equilibrium is characterized by price dispersion and consumer

search. The intensity of this search is related to the expected production cost

through its effect on price dispersion. Even though the level of the marginal cost

does not affect the trade-offs faced by the firms when setting prices, it alters

the range over which firms can choose those prices. In other words, the relative

benefits and costs of attracting the informed consumers are the same under low

and high costs. But, as production cost increases, the gap between the monopoly

price and the minimum profitable price (p∗) decreases (the extreme case being

c = υ).22 This implied negative relationship between price dispersion and pro-

duction cost induces consumers to search less when they expect high costs. This

can be seen in (1.9). The gains from search E [p− pmin|µ] are reduced as the

probability of high cost α increases. Thus, the indifferent consumer has a lower

22This is true for the case of consumers having downward slopping demands as long as theabsolute mark-up of a monopolist decreases with the marginal cost. See Appendix B for details.

14

search cost (equation 1.10) and the equilibrium search intensity decreases with

α. The following lemma summarizes this result and is central for the findings in

next section.

Lemma 2 Search intensity decreases when consumers expect higher production

cost: ∂µ∂α< 0

Proof: See Appendix A.

As long as the demand is composed of informed and uninformed consumers,

a market equilibrium implies price dispersion. This is not a result driven by the

heterogeneity in search costs. The last part of this section is devoted to extend the

results above to the case where g is degenerate and nonshoppers are homogeneous

in their search cost (si = s). Intuitively, when the search cost is sufficiently high,

the market equilibrium involves only shoppers searching.23 For very low search

cost, the gains from search are higher than its costs and everyone would want to

search. But we know that the competitive outcome implies no price dispersion

so it must be that if non shoppers are searching in equilibrium, they are doing

it with some probability q < 1. In order to analyze the equilibrium properties

better, let the number of shoppers be high or low; and the search cost be high,

moderate or low:

Definition 1 The number of shoppers λ is low (high) if λ is ≤ (>) than µ.

Given λ, search costs are defined to be low if s < E [p− pmin|µ = λ] , moderate if

E [p− pmin|µ = λ] ≤ s ≤ E [p− pmin|µ = µ] , and high if s > E [p− pmin|µ = µ].

23The existence of an atom of shoppers is enough to eliminate the Diamond Paradox (Dia-mond, 1971) where firms charge the monopoly price and consumers don’t search because thereis no price dispersion.

15

(a) High search cost (b) Low search cost

(c) Moderate search cost

Figure 1.2: Market equilibrium with shoppers and homogeneous nonshoppers

16

Figure 1.2 shows all possible equilibria. There is always a market equilibrium

with only shoppers searching (µ = λ) if E [p− pmin|µ = λ] < s. That is, when

the gains from search if only the shoppers do so are lower than the search cost of

the nonshoppers (Figure 1.2a and 1.2c). The rest of the equilibria imply search

by all types of consumers (µ > λ) and the search intensity is determined by the

roots q (if they exist) in

E [p− pmin|µ = λ+ (1− λ)q] = s (1.11)

This possibility arises if there is a low number of shoppers and the search cost

is low or moderate (Figure 1.2b and 1.2c). In the case of moderate search cost

there are two equilibria where µ > λ. The equilibrium with the smaller root q

is unstable, while the other is locally stable (as well as the one with q = 0).

Table 1.1 and the following corollary to Proposition 2 summarize the equilibrium

results.

Corollary 1 There can be one, two or three possible market equilibria when there

are λ shoppers and (1− λ) consumers with homogeneous search cost s > 0

If search cost is high: equilibrium is unique and µ = λ

If search cost is low: equilibrium is unique and µ > λ

If search cost is moderate

and the number of shoppers is high: equilibrium is unique and µ = λ

and the number of shoppers is low, there are three equilibria: i) µ1 = λ,

ii) µ2 > λ and iii) µ3 > µ2 > λ.

Similarly to the case of heterogeneous search costs, consumers have less in-

centive to search if they expect higher production costs. However, the market

17

Shoppers (λ)

Search Cost (s) low high

low µ > λ µ > λ

moderate µ = λ, µ1 > λ, µ3 > µ2 > λ µ = λ

high µ = λ µ = λ

Table 1.1: Market equilibrium with shoppers and homogeneous nonshoppers

search intensity only changes with α if the initial equilibrium involves searching

by nonshoppers. When consumers expect higher production costs they search less

since higher cost implies lower gains from search.24 The importance of this will

be seen in the next section in which I present a dynamic setup where consumers’

priors are based on past cost realizations.

1.3 Dynamics and asymmetric pricing

In this section, I present a simple dynamic model that parses out the conditions

under which asymmetric pricing in competitive markets holds. The main result

is captured by Proposition 3: firms react differently to positive cost shocks than

to negative shocks as long as those shocks are not iid. When search decisions are

linked to past cost realizations, firms face demands with different elasticities de-

pending on whether the cost dropped or rose in the past period. Different demand

elasticities are associated with different search intensity and imply asymmetric

cost pass-through by the firms. Before getting to the model setup, I present a brief

summary of how asymmetric pricing is defined and estimated in the literature.

Asymmetric pricing refers to the case where output prices react differently

24This is not the case in the unstable equilibrium that emerges when the number of shoppersis low search cost is moderate.

18

according to whether input prices have positive or negative changes. There is an

abundant empirical literature that suggests that asymmetric pricing is more the

norm than an anomaly. In particular, most studies find that prices react faster to

positive than to negative cost shocks (rockets and feathers pattern).25 In general,

most tests of asymmetric pricing estimate a dynamic error-correction model of

the following type:

∆yt =m∑

i=0

β+i (∆xt−i)

+ +m∑

i=0

β−i (∆xt−i)− + γ (yt−1 − δ0 − δ1xt−1) + εt (1.12)

where yt and xt represent output and input prices, and ∆ their change with

respect to the levels in the previous period. The model in (1.12) allows for

different effects of positive and negative cost shocks on prices, and assumes that

the output price adjusts completely to a cost shock after m periods. The last

term in parenthesis is the error-correction-term that accounts for the current

deviations from a long-run equilibrium relationship between the output and input

prices. Hence, the parameter γ is expected to be negative.

By separating the effects of positive and negative cost changes, a cumulative

response function (CRF) can be constructed for each type of shock. A CRF

predicts the amount of the price adjustment completed after k periods from

a one-time cost shock. Evidence of the rockets and feathers would consist on

the CRF identified with positive shocks being greater than the one for negative

shocks. If both cumulative functions are plotted against the number of periods

away from the cost change, we would expect the difference to be important in

the first periods after the cost changed and disappear as we approach to m.26

25See footnote 3 in the introduction for references on empirical work.26In general, data restrictions prevent the econometrician from including a sufficient num-

19

A simple model can be used to explain the rockets and feathers pattern.

Consider a dynamic environment where the static game presented in the previous

section is repeated over time. Assume that at the beginning of each period, nature

chooses a high or low production cost with probabilities α and (1− α). After

that, each firm observes the cost realization and sets prices while consumers

observe the previous period cost realization and decide whether to search or not.

Once the market clears, Nature draws another production cost and the process

is repeated. Since the main motivation for this model is to explain asymmetric

pricing in markets with atomistic firms, I ignore the possibility of collusion among

firms.

There are two sources of price variation over time in this setup. On the one

hand, prices can change as a reaction to a change in the production cost. All else

equal, a higher production cost implies higher expected prices in the market. But

on the other hand, market prices can vary as a result of a change in consumers’

priors. This is an indirect effect on prices that materializes through the variations

on consumers’ search intensity. Firms can anticipate this change in the search

intensity and adjust prices accordingly.

The expected market prices are completely characterized by the current pro-

duction cost level and the amount of search in the market. For simplicity, let the

probability of high costs follow a Markov process α = h(ct−1) where h (cH) = ρ

and h (cL) = (1− ρ) with 0 < ρ < 1. It then follows that there is a one-to-one

map between the previous period cost and the actual search intensity. Therefore,

the state of the economy can be represented by past and current cost realizations.

Denote the current state by k = (ct−1, ct). Since production costs can only be

ber of lags in (1.12) such that the CRF is estimated for all the periods it takes the price toaccommodate to the cost change (Peltzman, 2000).

20

low or high, the set of possible states is given by the set K = LL,LH,HL,HH

with ki = K (i). Given a current state ki, the probability of moving to a new

state kj next period is denoted by the element Pij in the following transition

matrix:

P =

ρ 1− ρ 0 0

0 0 1− ρ ρ

ρ 1− ρ 0 0

0 0 1− ρ ρ

(1.13)

Thus, if the current state involves low actual and low past cost realizations

(k1 = LL), it can never happen that the next state indicates high as the previous

cost (P13 = P14 = 0). Last, there is a unique invariant distribution for K and is

represented by π = ρ/2, (1− ρ) /2, (1− ρ) /2, ρ/2 .

In this simplified world, it takes only two periods for prices to fully adjust to

an isolated cost change. After a shock, firms increase (decrease) prices reacting

to bigger (lower) production costs. In the following period, assuming marginal

cost does not change, firms adjust prices to be consistent with the new updated

prior used by consumers. After two periods, the prices are in line with the new

cost level, and the size of the price adjustment is the same, independent of the

sign of the cost shock.27 Therefore, asymmetric pricing, if any, has to be observed

in the first period of adjustment to a cost shock.

We are interested in finding the conditions such that β+0 6= β−0 in (1.12).

First, consider β+0 and denote pk as the average market price when the state of

the economy is k. For a positive cost shock to occur, the previous cost realization

has to be low. Thus, the previous state was either LL or HL and the new state

27Moving from a state LL to HH implies the same price change than moving from HH toLL.

21

is LH. Similarly for β−0 ; the state of the period in which the cost drops can only

be HL while the previous state could have been either HH or LH. The expected

change in prices to a positive and negative cost shock are, respectively:

E

[4p4c+

]= Pr (HL) Pr (LHt|HLt−1) [pLH − pHL] +

Pr (LL) Pr (LHt|LLt−1) [pLH − pLL] (1.14)

E

[4p4c−

]= Pr (LH) Pr (HLt|LHt−1) [pLH − pHL] +

Pr (HH) Pr (HLt|HHt−1) [pHH − pHL] (1.15)

and using the transition and unconditional probabilities (P and π), the difference

becomes

E

[4p4c+

]− E

[4p4c−

]=−1

2ρ (1− ρ) [(pHH − pHL)− (pLH − pLL)] (1.16)

This last equation summarizes the conditions for asymmetric pricing. Note

that the economy can not move from a state HL to a state HH, so pHH − pHL

represents the change in expected prices after an increase in production cost

holding consumers’ priors at α = ρ. Likewise, pLH − pLL represents the increase

in prices if consumers’ priors are α = 1− ρ. other words, β+0 6= β−0 if the the cost

pass-through is sensitive to the priors held by consumers, and those priors are

not iid (ρ = 1/2).

Another way of seeing the drivers behind asymmetric pricing is by decompos-

ing (1.16) into: i) The effect of past cost on consumers’ priors, ii) the effect of

those priors on the search intensity, and iii) the effect of the search intensity on

the cost pass-through. That is, (1.16) can be approximated by

E

[4p4c+

]− E

[4p4c−

]≈ −1

2ρ (1− ρ) |∆c| ∂2pt

∂ct∂ct−1

=

=−1

2ρ (1− ρ) |∆c| ∂

2pt

∂ct∂µ

∂µ

∂α

∂α

∂ct−1

(1.17)

22

In the previous section, Lemma 2 showed that a higher expected production cost

generates less search by consumers. Lower gains from search are associated with

higher costs since, as the gap between the marginal cost and the monopoly price

is reduced, price dispersion decreases. Thus, the equilibrium pool of informed

consumers µ decreases with α. This is also true when g (·) is degenerated and the

equilibrium involves searching from nonshoppers (Corollary 1) as the probability

of a nonshopper searching increases (q (s > 0, α′) > q (s > 0, α′′) with α′ < α′′).28

If only shoppers are searching, the change in priors affects the benefits from search

but it might not be enough to induce nonshoppers to search (q = 0).

Now turn to the pass-through effect. An increase in the amount of informed

consumers is similar to an increase in the expected demand elasticity faced by

each firm. The limiting cases of perfect competition and monopoly are useful

benchmark cases. In a perfectly competitive environment, prices are driven en-

tirely by costs and a complete pass-through is expected after a cost shock. This

is not the case for a monopolist where the interaction between the demand and

cost determines market prices. In the case of consumers with homogeneous unit

demands, a monopolist sets prices independently of the cost level and the corre-

sponding pass-through is zero. Other assumptions on the demand function (linear

or constant elasticity, for example) allow for positive pass-through but still lower

than one.29

From the previous analysis, it can be inferred that as the number of informed

consumers increases, the market becomes more competitive and the link between

28A potential unstable equilibrium is ignored.29For demand functions where the monopolist pass-through is greater than one, the gap

between monopoly price and marginal cost increases with c. Since this implies that consumerssearch more when cost increases, the combined effect of search intensity and cost pass-throughdoes not change.

23

costs and prices is stronger. In other words, firms compete more fiercely for the

increasing mass of informed consumers by setting prices closer to marginal cost.

As a result, the cost pass-through is expected to increase with µ.

The expected market price for a given cost realization c and prior α is given

by

E [p|c] = υ −υ∫p∗

F (p, c)dp (1.18)

where the price distribution F (·, c) is the market equilibrium distribution (F (·, c, µ)

in (1.8) with µ = λ+(1− λ) g (s) from (1.10)). Integrating by parts and deriving:

∂E (p|c)∂c

= 1− (1− µ)

2µlog

[1 + µ

1− µ

](1.19)

The pass-through effect is positive for any value of µ. Using L’Hopital rule, it

can be checked that µ = 1 implies a complete pass-through while if µ = 0 there

is no price adjustment.30 The derivative of (1.19) with respect to µ confirms that

the cost pass-through is higher as the market becomes more competitive.

Combining (1.19) and the fact that higher priors generate less search (Lemma

2), the sign of the asymmetry in (1.17) is determined by the process behind α.

The next proposition summarizes the result.

Proposition 3 Asymmetric pricing occurs if cost is not iid. Moreover, prices

rise faster than they fall under cost persistence (ρ > 1/2).

Proof: See Appendix A.

30Note that the response of prices to production costs doesn’t depend on consumers’ reser-vation price υ. This is important when analyzing the case of sequential search by consumers.Any equilibrium that involve firms setting low prices such that consumers prefer to buy insteadof keep searching will not generate asymmetric pricing.

24

To summarize, asymmetric pricing occurs as a result of changes in the demand

faced by each firm when cost increases than when it decreases. In the case of

rockets and feathers, firms face a more inelastic demand if the marginal cost drops

than when it goes up. Suppose that marginal cost is currently high, consumers

expect it will remain high, so they expect little price dispersion and search very

little. If in fact, marginal cost drops, firms have few incentives to lower their prices

because consumers aren’t searching very much. On the other hand, if marginal

cost is currently low, it is likely to stay low, so next period’s price dispersion is

expected to be high, consumers search increases, and the response to a positive

cost shock is to pass most of it to prices.

An empirical implication of this model is that price dispersion generated by

costly consumer search is present at all times. Other models that have been

suggested to explain asymmetric pricing imply firms playing pure strategies most

of the time (see Lewis (2003) and Borenstein et al. (1997)). This feature is

analyzed in the retail gasoline market in Chapter 2. In the next section, I extend

the results to markets with more than two firms.

1.4 More sellers

In this section, I extend the results of sections 2 and 3 to atomistic markets.

The setup of the model is the same as the one presented above with the only

exception that the number of firms n is allowed to be greater than two. The

reason to present the results in a separate section is that I need to use simulations

to characterize the equilibrium since the Nash Best Response for the firms become

less tractable when as n > 2.

I again start by analyzing the firms’ NBR of the static game. With more

25

sellers in the market, the proportion of uninformed consumers that buy from

each seller decreases. This lowers the expected profits per firm. At the same

time, there are more firms disputing the mass of informed consumers. Thus, if

a firm wants to charge the lowest price in the market, it has to set lower prices

the larger the number of stores is. Restating equations (1.3) to (1.7) to account

for n > 2, and solving (1.7) one can find the unique symmetric equilibrium for

the firms. Given consumers’ search intensity and marginal cost, the NBR implies

firms pricing from the following cdf:

F (p, c, µ) = 1−(

(1− µ)(υ − p)

nµ(p− c)

) 1n−1

(1.20)

with support[c+ (1−µ)(υ−c)

1+(n−1)µ, υ]. The proof of Proposition 1 (in Appendix A) is

done for n > 2 and follows Varian (1980).

The changes in F (·) are plotted in Figure 1.3. The presence of more stores in

the market increases the likelihood of setting prices in the extremes of the distri-

bution. This is because the chances of being the lowest price in the market de-

crease with n and middle-range will never be enough to capture the informed con-

sumers. But the strengthening of the business-stealing and surplus-appropriation

effects is not symmetric. As n increases, the probability of being the lowest price

in the market decreases exponentially while the benefits from charging high prices

decrease at a rate 1/n. Thus, the surplus-appropriation effect becomes relatively

more important than the business stealing effect and firms prefer to increase the

likelihood with which they set prices close to the monopoly price than on low

prices.

As the number of sellers increase, the cdf becomes flatter over low and medium-

range prices and the expected price in the market increases. In the limit, the price

26

Figure 1.3: Equilibrium price distribution and number of stores (c = 0, µ = 0.2)

distribution converges weakly to the monopoly price (Stahl, 1989; and Janssen

and Moraga-Gonzlez, 2004). Nevertheless, the support of the price distribution

increases with n and its lower bound approaches marginal cost. That is, there is

always a positive probability (for consumers) of finding very low prices.

In a market equilibrium, consumers decide endogenously their optimal search-

ing strategy. The effect of the number of sellers on the equilibrium search intensity

is determined by the effect of n on the expected price and expected minimum

price. As in (1.9), the expected gains from search are now:

Ec [E [p− pmin|c, µ, n]] = Ec

v∫p∗

p (υ − c)(1− n [1− F (p)]n−1)

(n− 1) (p− c) (υ − p)[1− F (p)] dp

(1.21)

It was claimed above that the expected price increases with n. Intuitively, the ex-

pected minimum price decreases with the number of sellers since the lower bound

of the distribution support approaches the marginal cost. Therefore, consumers

have more incentives to search in more atomistic markets than in duopolies.

Proposition 4 Search intensity increases with n :

E [p− pmin|c, µ, n+ 1] > E [p− pmin|c, µ, n]

Proof: See Appendix A.

27

There are various ways to think about how competitive the market becomes

when the number of sellers increases. As n grows, prices approach the monopoly

price, but at the same time profits vanish. Furthermore, holding constant the

number of firms, a larger number of informed consumers implies a more elastic

demand faced by each firm. As µ increases, the market is more competitive and

prices decrease regardless of the number of firms. From (1.20), F (·, µ′) > F (·, µ′′)

if µ′ > µ′′.

The expected gains from search is a continuous function of µ, and -as with

n = 2- it is zero when µ = 0 (monopoly) or µ = 1 (perfect competition) and

increases as µ is away from those extremes. The conditions for unique market

equilibrium in Proposition 2 are related to the concavity of the gains from search.

Unfortunately, for markets with n > 2, the expression in (1.21) becomes less

tractable and I need to rely on simulations to show its concavity. Table 1.2 shows

the numerical values for E [p− pmin|c, µ, n] as a function of different combinations

of marginal cost values, amount of informed consumers, and number of firms in

the market. It can be seen that the gains from search increase with µ at an

increasing rate, reach a maximum and then decrease toward zero. The plots in

Figure 1.4(a) represent the first panel of Table 1.2 and confirm the concavity

assumption. Lastly, the effect of n on the amount of informed consumers that

maximizes the expected gains is shown in Table 1.3.

With concavity guaranteed, Proposition 2 can be applied to the case of more

atomistic markets. Given the production cost and consumers’ priors, there is a

unique market equilibrium that is characterized by price dispersion and active

search by consumers. Consumers search because they expect price dispersion, and

firms generate price dispersion because consumers are searching. The amount of

search in equilibrium is influenced by the expectations over the marginal cost.

28

υ/c = 2 υ/c = 5 υ/c = 10

µ|n 10 50 100 10 50 100 10 50 100

0.1 0.268487 0.6282 0.75255 0.42958 1.005121 1.20408 0.483277 1.130761 1.35459

0.2 0.396569 0.744037 0.838856 0.634511 1.19046 1.34217 0.713825 1.339267 1.509941

0.3 0.472803 0.796879 0.875598 0.756486 1.275007 1.400956 0.851046 1.434383 1.576076

0.4 0.522712 0.827401 0.896187 0.83634 1.323841 1.4339 0.940882 1.489322 1.613137

0.5 0.556532 0.846985 0.909214 0.890452 1.355158 1.454743 1.001758 1.524553 1.636586

0.6 0.578868 0.860017 0.917898 0.926188 1.376027 1.468637 1.041962 1.54803 1.652216

0.7 0.591375 0.868404 0.923613 0.946199 1.389447 1.477781 1.064474 1.563128 1.662503

0.8 0.592896 0.872529 0.92676 0.948634 1.396046 1.482815 1.067214 1.570552 1.668167

0.9 0.575174 0.870377 0.92638 0.920279 1.392604 1.482208 1.035314 1.566679 1.667484

Table 1.2: Expected gains from search

(a) Number of firms (υ/c = 2) (b) Cost variation (n = 100, υ = 2)

Figure 1.4: Maximum expected gains from search and the number of firms

n 2 102 202 302 402 502 602 702 802 902 1002

bµ (n) 0.6349 0.8471 0.8626 0.8704 0.8755 0.8792 0.882 0.8844 0.8863 0.888 0.8894

Table 1.3: Maximum E[p− pmin|c, n] and n

29

Note that when marginal cost is high, the expected price, as well as expected

minimum price, increase. Since the latter effect is stronger than the former

(see Lemma 2), the expected price dispersion in the market decreases with the

marginal cost. This is shown in 1.4.b for parameter values n = 100 and υ = 2.

The last step needed for the asymmetric pricing and rockets and feathers

results is to show that the pass-through increases with the amount of search by

consumers. That is, ∂2E(p)∂c∂µ

> 0 in (1.17). For the reasons explained above, it is

expected that for a given level of search, the pass-through in a duopoly is bigger

than in a market with more firms. Start assuming that µ = 0. In this case, each

firm is a monopolist over half of the consumers in the market. The pass-through is

zero independent of the number of firms. But as consumers become informed, the

surplus-appropriation effect is stronger in more atomistic markets. That is, firms

prefer high prices to low prices, and average prices are further from the marginal

cost as the number of firms increases. The fact that in atomistic markets each

firm is more concentrated on its captive consumers explains why the incentives to

adjust prices to cost changes are lower. In Figure (1.5), the pass-through effect

is drawn for markets with different numbers of firms and parameters υ = 2 and

c = 0. The pass-trough approaches 1 as the proportion of informed consumers

dominates the market, but for n > 2, this convergence occurs only when the

market is very close to perfectly informed.

To conclude, the rockets and feathers result can be extended to markets with

more than n firms since all the conditions found in the duopoly hold. Namely: i)

consumers search less if they expect a higher cost, and ii) the cost pass-through

by firms increases with the amount of informed consumers. Under persistence in

the cost shocks, the asymmetric pricing takes the form of the rockets and feathers

pattern.

30

Figure 1.5: Cost pass-through

1.5 Conclusion

This paper develops a model that explains the widely observed rockets and feath-

ers price pattern. The model links the firms’ asymmetric response to cost shocks

to the fact that consumers are imperfectly informed about both market prices and

the industry’s production cost. Consumers’ search decisions affect the elasticity

of the expected demand faced by firms and therefore their cost pass-through.

If production cost shows serial correlation, the number of informed consumers

in the market depends on the previous cost realization. As a result, the cost

pass-through exercised by firms is different when the cost drops than when it

raises.

The simplicity of the model helps to identify the forces behind asymmetric

pricing. The assumptions on both the cost process and consumers’ learning of

production cost could be modified to better approximate the quantitative prop-

erties of the observed rockets and feathers pattern in each market.

Contrary to public opinion and previous work suggesting that collusive be-

havior was the cause behind asymmetric pricing, this paper shows that it can well

be the outcome of a competitive market. This finding reinforces the importance

31

of consumer search models in explaining actual markets functioning. The next

Chapter of this dissertation is a step in that direction.

32

Appendix A - Proofs

Proof of Proposition 1. This proof is done for the n firms case since it is also

used in Section III. Therefore, p∗ = c+ (1−µ)(υ−c)1+(n−1)µ

in (1.6).

To show that F (·, c, µ) is a unique symmetric NR the proof is divided in

three steps (to simplify notation, ignore the fact that F is conditional on (c, µ)).

First, it shows that there are no point masses in the equilibrium pdf . Second, for

ε > 0, F (p∗ + ε) > 0 and F (υ − ε) < 1. Last, there are no gaps in the support

of F (p)

1. Assume there exist a price p ∈ (p∗, υ] such that Pr (p = p) ≡ F (p) > 0

(by definition, F (p∗) = 0). Then, there is an arbitrary small ε such that

F (p− ε) = 0.A firm could deviate from F (·) by applying F d (·) similar to

F (·) with the exception that F d (p) = 0 and F d (p− ε) = F (p) . The

expected gains for the deviator can be decomposed to four scenarios, depending

on the prices charged by the other firms. Let pl be their lowest of the n prices in

the market. If pl < p− ε :

n−1∑j=1

(n− 1

j

)F (p− ε)j [1− F (p− ε)]n−1−j

−(1− µ)

nε

(1.22)

If pl > p :

−ε(

(1− µ)

n+ µ

)[1− F (p)]n−1 (1.23)

When pl = p :

n−1∑j=1

(n− 1

j

)F (p)j [1− F (p)]n−1−j

µ

(1− 1

j

)(p− c)−

((1− µ)

n+ µ

)ε

(1.24)

33

Lastly, if pl ∈ (p− ε, p) , the expected gains are:

n−1∑j=1

(n− 1

j

)[F (p)− F (p)− F (p− ε)]j [1− F (p)]n−1−j

µ (p− c)−

((1− µ)

n+ µ

)ε

(1.25)

As ε→ 0, (1.22) and (1.23) go to zero while (1.24) and (1.25) remain positive.

2. Suppose F (υ − ε) = 1. Then at setting p = υ generates an increase in

profits (with respect to υ−ε) and no loss in customers. Similarly, if F (p+ ε) = 0,

it has to be that π (p+ ε) = π (υ) . By charging p = p + ε/2, profits are bigger:

π (p+ ε/2) > π (p∗) = π (υ) .

3. Suppose there exists an interval (p1, p2) such that F (p1) = F (p2) . Then,

by placing some density on p ∈ (p1, p2) , a firm will gain by increasing its markup.

There is no expected loss since by part 1 of the proof, there are no ties at p1.

Given, 1, 2, and 3 above, the only function that satisfies

πs(p)(1− F (p))n−1 + πf (p)F (p) = π(υ)

is:

F (p) = 1−(

(1− µ)(υ − p)

nµ(p− c)

) 1n−1

Proof of Lemma 1. I first show that there exists a unique global maximum

µ for E [p− pmin] and strict local concavity around E [p− pmin|µ = µ] . Then,

concavity everywhere is provided. From (1.9),

∂E[p− pmin]

∂µ=

(υ − c) (2− µ)

2µ3

2µ (2 + µ)

(2− µ) (1 + µ)− log

[1 + µ

1− µ

]with Lim

µ→0

∂E[·]∂µ

→ υ−c3

and Limµ→1

∂E[·]∂µ

→ −∞. The term in curly brackets determines

the sign of this expression. Critical points are at µ = µ 6= 0, 1,

log

[1 + µ

1− µ

]=

2µ (2 + µ)

(2− µ) (1 + µ)(1.26)

34

At µ = 0, LHS = RHS. The difference in slopes between RHS and LHS is:

∂LHS

∂µ− ∂RHS

∂µ= − 4µ2 (1− 2µ)

(1− µ) (2 + µ (1− µ))2

which is positive (negative) for µ < (>) 1/2. Since at µ = 1, LHS > RHS, there

is a unique critical point at µ > 0.5.31

The second derivative of (1.9) is:

∂2E[p− pmin]

∂µ2=

− (υ − c)

(1− µ) (1 + µ)2 µ4

2µ (3 + µ (2− µ (3 + µ)))

(3− µ) (1− µ) (1 + µ)2 − log

[1 + µ

1− µ

]Using (1.26) and rearranging, at µ,

∂2E[p− pmin]

∂µ2=

2 (υ − c) µ3 (1− 2µ)

(1− µ) (2− µ) (1 + µ)2 µ4< 0

For concavity everywhere,

2µ (3 + µ (2− µ (3 + µ)))

(3− µ) (1− µ) (1 + µ)2 ≥ log

[1 + µ

1− µ

]At µ = 0, both expressions are equal to zero. For µ > 0, it can be verified that

∂LHS∂µ

> ∂RHS∂µ

> 0

Proof of Proposition 2. Reexpress (1.10) using (1.9)

(υ − E [c])(1− µ)

2µ2

[log

[1 + µ

1− µ

]− 2µ

]= g−1(

µ− λ

1− λ)

At µ = λ+ (1− λ) g (0) , the RHS is zero while the LHS is positive. By Lemma

1, LHS is concave and lower than υ. Thus, g−1 cuts from below the expected

gains from search at least once. If λ > µ, it is easy to see that there is a unique

solution to (1.10). If λ < µ, the possibility of multiple solutions is eliminated if

g−1 has steeper slope than the LHS for any value of µ in the range (λ, µ)

31Numerically, the maximum can be shown to be µ ≈ 0.634816

35

Proof of Lemma 2. Let the equation in (1.10) be represented by G. Using

(1.9):

G = (υ − E [c])1− µ

2µ2

[log

[1 + µ

1− µ

]− 2µ

]− g−1

(µ− λ

1− λ

)where µ = λ+ (1− λ) g (s) . Then, by the IFT,

∂s

∂α= −

∂G∂α∂G∂es

The numerator is negative since α increases E [c]. The denominator is

∂G

∂s= (1− λ)

∂g

∂s

(∂E [p− pmin|c, µ]

∂µ− ∂g−1

∂µ

)< 0

Since at s the inverse cdf cuts the expected price differential from below, the

term in parenthesis is negative.

The same argument applies to the case of degenerate g (·). E [p− pmin|µ = λ+ (1− λ) q] =

s could have one or two roots q depending on the size of λ and s. The stable equi-

librium has E [·] cutting s from above. As α increases, E [·] gets flatter and q

(hence µ) decreases

Proof of Proposition 3. If cost is iid consumers would not update priors

( ∂α∂ct−1

= 0) and there is no asymmetric pricing in (1.17). When cost is persistent,

h (cH) > h (cL) so ∂α∂ct−1

> 0 and ρ > 1/2. The derivative of the pass-through

(1.19) w.r.t. µ∂2E (p|c)∂c∂µ

=1

2µ

[log

[1 + µ

1− µ

]− 2µ

(1 + µ)

]is positive since log

[1+µ1−µ

]> 2µ. Therefore, ∂2pt

∂ct∂µ(+)

∂µ∂α(−)

∂α∂ct−1

(+)

< 0 and E[4p4c+

]−

E[4p4c−

]> 0 in (1.17)

Proof of Proposition 4. As long as the conditional gains from search increase

with n, ∂es∂n> 0 in (1.10) and ∂q

∂n≥ 0 in a stable equilibrium of (1.11). The gains

36

from search are:

E [p− pmin|c, n] =

v∫p∗

pn [1− F (p)]n−1 f (p) dp =

v∫p∗

p(v − c)

(n− 1) (p− c) (v − p)n [1− F (p)]n dp

Define z = 1−F (p). Then, p = υ(1−µ)+cnµzn−1

(n−1)(p−c)(υ−p)and dp = − (1−µ)µn(n−1)(υ−c)zn−1

(z(1−µ)+µzn)2dz.

Changing variables,

E [p− pmin|c, n] =

1∫0

nzn−1

[υ(1− µ) + cµnzn−1

(1− µ) + µnzn−1

]dz = υ′

1∫0

nzn−1

1 + µ(1−µ)

nzn−1dz

wlg, the marginal cost can be normalized to 0 and υ adjusted to υ′. Define

An+1 = 1 + µ(1−µ)

(n+ 1) zn and An = 1 + µ(1−µ)

nzn−1:

E [p− pmin|n+ 1]− E [p− pmin|n] = υ′1∫0

(n+ 1) zn

An+1

− nzn−1

An

dz =

= υ′1∫0

zn−1µ/ (1− µ) [n− (n+ 1) z]

An+1An

dz =

= υ′n/(n+1)∫

0

zn−1µ/ (1− µ) [n− (n+ 1) z]

An+1An

dz − υ′1∫

n/(n+1)

zn−1µ/ (1− µ) [(n+ 1) z − n]

An+1An

dz ≥

≥ υ′[1 + µ

(1−µ)(n+ 1)

(n

n+1

)n] [1 + µ

(1−µ)n(

nn+1

)n−1] 1∫

0

zn−1 µ

(1− µ)[n− (n+ 1) z] dz = 0

37

Appendix B - Multiunit Demands

The model used in this paper assumes that consumers have unit demands. In

this Appendix, I study the robustness of the results with respect to various de-

mand assumptions. As it will become clear below, the characterization of the

static equilibrium is not itself altered, but both the comparative statics of price

dispersion with respect to the firms’ production cost, and the expected cost pass-

through with respect to the search intensity in the market are sensitive to the

demand functional form. From (1.17) we know that these two effects are im-

portant in determining the existence of the rockets and feathers pattern since

sign

(E

[4p4c+

− 4p4c−

])= sign

(−∂µ∂α

∂2pt

∂ct∂µ

)(1.27)

With unit demands, consumers search less if they expect higher production

costs ( ∂µ∂α

< 0). The firms’ cost pass-through increases as the market becomes

more competitive ( ∂2pt

∂ct∂µ> 0). In what follows, I start by analyzing the duopoly

Nash Best Responses for a general demand function and then restrict the anal-

ysis of price dispersion and search intensity to the cases of linear and constant

elasticity demands.

Assume all consumers demand the final good according to the demand func-

tion q (p) with q′ < 0, ε = −∂q∂p

qp, and a unique price pm = arg max

pq (p) (p− c) .

Let ψ (p) = q (p) (p− c) denote the profit per consumer for a firm that sets a price

p. With multiunit demands, consumers not only decide whether to search or not,

but also the number of units they’ll buy once they observe a price. Thus, from the

point of view of the firms, there are two price elasticity measures (ε and µ) since

there are two channels by which setting a lower price increases the expected ag-

gregate demand. First, the probability of attracting µ more consumers is greater,

38

and second, each consumer that shows up in the store buys more units.

By the same argument used in Proposition 1, as long as there is a positive

mass of informed and uninformed consumers, there is no pure strategy equilib-

rium and the unique mixed strategies equilibrium implies firms drawing prices

from distribution function F (·). By charging any price in F ’s support a firm

obtains the same expected profit as if they were setting the monopoly price. The

equivalent to Equation (1.7) is:

π (pm) =(1− µ)

2ψ (p)F (p) +

(1 + µ)

2[1− F (p)]ψ (p)

substituting ψ (p) and π (pm) and reorganizing

F (p) =(1− µ)

2µ

[(1 + µ)

(1− µ)− q (pm) (pm − c)

q (p) (p− c)

](1.28)

F (p) is concave since ψ (p) is concave over the relevant range of prices (ψ′ (p) ≥ 0

for p ≤ pm) and its concavity increases with the elasticity of consumer demand

(ε).32 Both the surplus-appropriation and business-stealing effects explain the

equilibrium in mixed-strategies. But -as opposed to the case of unit demands,

when firms set prices below the monopoly price the loss of surplus is partially

compensated with the fact that the uninformed consumers buy more units at

lower prices. Thus, the support of the price distribution increases with ε. It

can be checked that with unit demands (1.28) becomes (1.8). Figure 1.6 shows

a comparison of the price distributions associated with three different demand

functions i) linear, ql = a − bp, ii) unit demand, qu = Ip<v, and iii) constant

elasticity, qce = K/pθ, θ > 1. The parameters a, b, v, θ,K were chosen such

that the monopoly outcome is identical for the three cases (Figure 1.6 .a), and

32This is not the case when there are more than 2 firms in the market.

39

a) Demand functions b) µ = 0.2

c) µ = 0.5 b) µ = 0.8

Figure 1.6: Consumer demand and firms’ NBR

the elasticity at the monopoly price is the same for i) and ii).33 As equation (1.8)

and Figure 1.6 makes visually clear, the differences in the price distributions are

attenuated as the number of informed consumers (µ) in the market increases.

To complete the characterization of the market equilibrium, we need to focus

next on the search decisions made by consumers, given the firms NBR. Consumers

are concerned about the price dispersion in the market since they would pay either

the minimum price or a random price depending on whether they choose to search

or not. Thus, given the number of informed consumers, the price dispersion in the

market is defined as the difference between the average price and the minimum

33The parameter values are: c = 1, v = 8, a = 2.14286, b = 0.1428, θ = 8/7 and K = 10.7672.

40

expected price. With linear demands, the average price is given by:

E [p|µ, c] = Ec

v∫p∗

pdF (·, c, µ)

=

=(1− µ)

4bµ

2 (a+ bE [c])

(1− µ)−

(a− bE [c])

[(2µ (1 + µ))1/2

(1− µ)+

Ω

2

](1.29)

And the minimum expected price is:

E [pmin|µ, c] = Ec

v∫p∗

p [1− F (p)] dF (·, c, µ)

=

=1

32bµ2

16(a+ bE [c])µ2 + (a− bE [c])

[(6− 14µ) (2µ (1 + µ))1/2 + 3 (1− µ)2 Ω

]where Ω = log [A/B], A = (1 + µ)1/2 − (2µ)1/2 and B = (1 + µ)1/2 + (2µ)1/2;

hence Ω < 0.

The difference simplifies to:

E [p− pmin|µ, c] =(a− bE [c])

32b

(1− µ)

µ2

[−6 [2µ (1 + µ)]1/2 − (3 + µ) Ω

](1.30)

In the extreme cases of µ = 0 (monopoly) and µ = 1 (perfect competition)

there is no price dispersion. It can also be checked that E [p− pmin|µ] is a strictly

concave function of µ (Lemma 1) and a unique market equilibrium exists under

mild conditions (Proposition 2). Similar to the case of unit demands, the price

dispersion has a linear and negative relationship with the production cost. The

reason for this is that with higher production costs, the trade-offs faced by the

firms when choosing prices are not affected, but the range of prices over which

firms choose from shrinks and the new price distribution is rescaled.34 The analog

extreme case to c→ v in the unit demand case is when c→ a/b.

34The bounds of the support are pm = a+bc2b and p∗ = (a+bc)

b − (a−bc)b

[µ

2(1+µ)

]1/2

with

∂(pm−p∗)∂c = − 1

2

(µ

1+µ

)1/2

< 0.

41

But the negative relationship between price dispersion and production cost

is not enough to characterize the new search intensity in the market. This is so

because the search decision by consumers with multiunit demands depends not

only on the gap between the mean and minimum prices (price dispersion) but

also on the actual price levels. A change in the cost of production has two effects.

First, as the cost increases, the gains from search per unit bought are lower (

1.30). Second, as cost (hence prices) increases, consumers plan to buy less units

and so their incentives to find a good deal are lower (less search). Therefore,

the search decision is made by comparing the search cost to the expected gain in

consumer surplus of paying the minimum price in the market instead of a random

price:

E[CS(pmin)− CS(p)|µ, c] =Ec

[(a− bc)2] (1− µ)

64bµ2A2r (µ)

where r (µ) = 16µ−2 (3 + µ) [2µ (1 + µ)]1/2−“1 + 3µ− 2 (2µ (1 + µ))1/2

”log

hA(7+µ)B(1−µ)

(1+µ)4

iis positive

and has the properties r (0) = 0 and r (1) = 0.

As expected, increases in the production cost generate less consumer search,

∂µ∂α< 0. In other words, when higher production costs are expected, the indifferent

consumer will have a lower search cost because a) she expects to buy fewer units,

and b) the gains from search per unit decrease. Figure (1.7) shows the total gains

from search under low and high cost as well as the difference between the gains

from search and the price dispersion (E[CS(pmin)−CS(p)|µ, c] and E[p−pmin|µ, c]

respectively).35

In order to obtain the rockets and feathers result, we need to verify that the

cost pass-through increases with the search intensity. This is done by differenti-

35The parameters values are: a = 10, b = 1, cH = 2, cL = 1

42

Figure 1.7: Gains from search and price dispersion with linear demands.

ating (1.29) twice:

∂2E [p]

∂c∂µ=

1

8µ2

[Ω− 2

(2µ

1 + µ

)1/2]> 0 (1.31)

Starting with the monopolist cost pass-through (∂pm

∂c= 1

2), as the search inten-

sifies, the pass-trough approaches the competitive outcome (full pass-through).

This result is related to the effect of expected production cost on price dispersion.

For some demand functions (linear and unit demand), the monopolist absolute

mark-up decreases with the production cost. This is in fact what makes the sup-

port of the price distribution to shrink (∂pm−p∗

∂c< 0) and lowers the gains from

search to consumers. At the same time, if the monopolist absolute mark-up de-

creases with the production cost, it must be that the monopolist cost pass-trough

is less than one (the competitive pass-through).

To summarize, demand functions for which the monopolist absolute mark-

up decreases with the cost of production generate the same qualitative effects

(equation 1.31) as the unit demand case and the rockets and feathers result is

guaranteed.

Contrary to the linear demand case, when the mopolist absolute mark-up in-

43

creases with c, higher production costs imply more price dispersion. For example,

when consumers’ demands is of the form qce = K/pθ, θ > 1. Following the same

argument as before, the price support (hence price dispersion) should expand

with the cost of production. Unless θ = 2, there is no explicit solution for the

lower bound of the price support. Using the IFT in (1.28),

∂p∗

∂c= −

∂F (p)∂c

∂F (p)∂c

∣∣∣∣∣p=p∗

= −p1+θ

(cθ

θ−1

)−θ(c− (p− c) p−θ (θ − 1)

(θc

θ−1

)θ)c (p− c) (θ − 1)

(1− ( θc

θ−1)1−θ

p−θ(p−c)θ

)∣∣∣∣∣∣∣∣p=p∗

using the fact that expected profits are the same for p∗ and pm and simplifying36

∂p∗

∂c=p∗

c> 1

∂ (pm − p∗)

∂c> 0

since p∗ < pm = cθθ−1

.

Together with the expansion of the price support, the gains from search per

unit increase with the expected cost of production (unlike the in the unit and

linear demand cases). On the other hand, since consumers expect to buy less

units when the production cost increases, the final effect on the gains from search

is not clear.

In what follows, I concentrate on the case of θ = 2 and provide numerical

results for different values of the price-elasticity. The total gains from search are

E[CS(pmin)− CS(p)|µ, c] =(1− µ)

32µ2E

[1

c

] [(3 + µ) log

[(1− µ)

1 + 3µ− 2C

]− 6C

]36 p∗−θ (p∗ − c) (1 + µ) = (1−µ)

θ

(cθ

θ−1

)1−θ

44

(a) Price dispersion (b) Gains from search

Figure 1.8: Higher production costs. Constant elasticity demands.

where C = (2µ (1 + µ))1/2 .37 It is easy to see that increases in the production

cost implies less search. That is, even though the price dispersion increases with

the cost of production, consumers choose to search less because the number of

units they expect to buy is smaller. Figure 1.8 shows this for θ = 2 and K = 1

although the same properties hold for different parameter values.

Given that consumers with constant elasticity demands search less when they

expect higher production costs ( ∂µ∂α> 0), the rockets and feathers result will only

hold if the cost pass-through increases as the market becomes more competitive

(see 1.27). But from the previous discussion we know that this is unlikely since

the monopolist’ pass-trough is higher than 1. In fact, the expected pass-trough

in the market decreases as the number of informed consumers increases:

∂2E [p]

∂c∂µ=

1

8µ2 (1− µ)2

[8µ2 + C (1− 5µ)− (1− µ)2 log

[(1− µ)

1 + 3µ− 2C

]]< 0

37CS (p) =

∞∫p

1xθ dx.

45

a) Price dispersion b) Gains from search

c) µ = 0.25 d) µ = 0.75

Figure 1.9: Numerical simulations for θ > 1

46

The same qualitative result is found for other values of θ (Figure 1.9) and this

implies feathers and rockets instead of rockets and feathers. However, the quan-

titative effects are of a smaller order than in the case of unit or linear demands

since both the unitary gains from search as well as the quantity demanded by

each consumer change in the same direction when a higher cost of production is

expected.

47

Appendix C - Sequential Search

In this appendix, I explore the implications of changing the search protocol used

by consumers from nonsequential to sequential search. A consumer that searches

sequentially observes (costlessly) a price at the first store she visits and then

decides whether to visit (costly search) another store or go ahead and buy at

the observed price. As discussed in Section 1.1, sequential search can be optimal

when the search technology does not allow for significant economies of scale in

the sample size.

All else equal, sequential search is more desirable than nonsequential search

if consumers are uncertain about production costs. After observing a price, con-

sumers are able to update their prior beliefs about the cost realization and there-

fore the distribution from which market prices are drawn from. As it is shown

below, the rockets and feathers result is not robust to the search protocol used by

consumers. The reason is that when consumers are either shoppers or nonshop-

pers with homogeneous search cost, the equilibrium implies that only shoppers

search. Thus, changes in the cost of production don’t change the search intensity

in the market. Drawing from this insight, a necessary condition for rockets and

feathers points toward the heterogeneity in non shoppers’ search costs.

The change in the search protocol used by consumers modifies the stage game

presented in Section 1.2. It is now a sequential-move instead of simultaneous

game. After nature draws a production cost c ∈ C, firms choose their prices

simultaneously and commit to them for one period. Once prices are set, each

consumer visits a random store and decides between buying at the price observed

p0 or searching (with a cost si) for a better price in the remaining store. A

consumer that decides to search buys from the store with the lowest price (perfect

48

recall). To simplify the analysis, assume that the distribution of search costs g (·)

is degenerate at s >> 0. Thus, the market demand is composed by λ shoppers

and (1− λ) consumers with homogeneous search cost s. 38

Consumers randomly choose the first store they visit. Keeping the notation

used in (1.1), a consumer i that visits a store j has conditional utility

uij (ai, a−i, p) = υ − ai [Min [p]− si] + (1− ai) pj (1.32)

Let µj be the proportion of the consumers with positive search cost(

1−λ2

)that

visits store j and decides to sample another price. Thus, given a price vector

p = (pj, p−j) , the number of consumers that visit the two stores is ψ = λ +(µj + µ−j

) (1−λ

2

)< 1. Firm j′s payoff becomes

πj (pj, p−j, a, c) = (pj − c)

[ψIpj<p−j +

1

2Ipj=p−j +

(1− λ

2

)(1− µj

)Ipj>p−j

]

A consumer strategy is a function qi (·, p0, si) ∈ ∆ (A) of the observed price

p0 ∈ P and search cost si ∈ 0, s. Since shoppers always visit both stores,

qi (·, p0, s = 0) = 1 (search) for any p0.39 Firm j’s strategy profile is represented

by all possible price distributions conditional on the cost realization: fjc (·) =

(fjc (p))p∈P,c∈C .40 A SBNE (symmetric bayesian Nash equilibrium) or market

equilibrium consists on a strategy profile σ =(σD, σF

)and consumers’ beliefs

α′(p0, σ

F)

p0∈Psuch that i) σD = (qi (·, 0) , qi (·, s))i∈ND is a best response to

38The equilibrium characterization of the nonsequential search case with degenerated g (·) isdone by Definition 1 together with Corollary 1.

39Note that an equilibrium where ψ = 1 is not feasible. If consumers visit both stores, thefirms compete aggressively on prices and the competitive outcome results p = c. But then,nonshoppers don’t want to search.

40For expositional reasons, the cost realization now is expressed as a subindex of the strategyprofile instead of an argument of it.

49

σF =(fc

(p, σD

))p∈P, c∈C

, ii) σF is a Nash Best Response (NBR) to σD, and iii)

α′ = Pr(cH |p0, α, σ

F)

is the Bayesian update of prior α for each observed price

consistent with σF .

An intuitive and desired property for the optimal search rule is that it pre-

serves the reservation price property (RPP). That is, consumers decide to search

only if the observed price is above a certain reservation price. In such a case, q

would be a step function with q (p0 ≤ v, s) = 0 (don’t search) and q (p0 > v, s) = 1

(search) where v ≤ υ is the reservation price. Given a certain distribution of

prices, consumers that sample a high price have more incentives to pay the search

cost and visit the second store since it is likely that they will sample a better

price. But the RPP is not guaranteed in models that involve consumers searching

sequentially from an unknown price distribution. A consumer that observes a low

price p0 might now consider that the cost realization for the firms was low and

decide to continue searching for lower prices. If she observes a high price, the

fact that high cost is more likely might make p0 a bargain. As it is shown below,

the RPP is satisfied if the cost shocks meet some requirements..

Assume that the equilibrium NBR for each production cost level are repre-

sented by FH (p, r)p∈[p∗H ,r] and FL (p, r)p∈[p∗L,r] with FL (p, r) > FH (p, r) for all p

and p∗H > p∗L. The parameter r is -for the moment- an exogenous upper bound

for the prices selected by the firms. The expected benefit of sampling one more

price for a consumer that observes p0 is the expected price at the remaining store,

50

conditional on it being lower than p0.

EG(p0, r) = α′p0∫p∗H

(p0 − p) fH (p, r) dp+ (1− α′)

p0∫p∗L

(p0 − p) fL (p, r) dp =

= α′ (p0)

p0∫p∗H

FH (p, r) dp+ (1− α′ (p0))

p0∫p∗L

FL (p, r) dp (1.33)

where the second equality is obtained integrating by parts. A monotonic EG

implies a unique (if there is one) reservation price v such that EG (v, r) = s. Two

effects take place after a price p0 is observed: a direct and a learning effect. The

former is the standard effect when a consumer samples from a known distribution.

High observed prices imply a higher probability of sampling a lower price next

time. The learning effect isolates the new information contained in p0 about the

unknown sampling distribution. Higher observed prices could be due to higher

costs of production hence the gains from search are lower. Since the learning and

direct effects go in opposite direction, EG can be non-monotonic and the RPP

violated.

As it is usual in the literature of optimal sampling from unknown distributions,

assumptions are made to preserve the reservation price property. When the

price distribution is exogenous, assumptions rely on F (·) (Rothschild, 1974).

If the price distribution is endogenously determined, assumptions on either the

production cost structure or the number of shoppers guarantee a weak learning

effect.41 I will assume that, for a given search cost s, the gap between the two

41For example, Benabou et. al. (1993) assume low correlation in firm’s marginal costs. Intheir model, firms have production costs composed of both an idiosyncratic (real) and a common(inflation) shock. A consumer that observes a low price knows that the firm has a low cost andneeds to infer the cost of the remaining producer in order to decide to sample another price.When costs have low correlation, the learning effect is dominated by the direct effect and EG ismonotonic.

51

possible production costs (cH − cL) is sufficiently low.42 That is, I assume the

learning effect to be smaller than the direct effect so higher prices imply increasing

gains from search. Given the assumption on the cost gap, consumer strategy

profiles are restricted to the type:

σD = (qi (·, 0) , qi (·, s))i∈ND (1.34)

qi (p0, 0) = 1 for all p0

qi (p0, s) =

1 for p0 > v

0 otherwise

Shoppers always search while nonshoppers only if the price is above their

reservation value v. As in the case of nonsequential search, the existence of an

atom of shoppers eliminates the possibility of a Single Price Equilibrium (Stahl,

1989). Furthermore, the upper bound for any price distribution is v instead of

the monopoly price υ.

Lemma 3 Firms’ NBR imply fc (v) = 0 and Fc (v) = 1.

Proof. Given non-shoppers’ reservation price v, assume that firms draw prices

from a cdf Fc (·) with upper bound p. It is not possible that p < v since a firm

could deviate and increase its revenues (keeping the same number of consumers)

by setting p = v. Suppose p > v. A firm that charges p = p would not capture

any customers while, with p = v the customers are at least (1−λ)2. Thus, the only

possibility left is p = v

A consequence of this lemma is that in any market equilibrium, consumers

with positive search cost never observe a price above their reservation value, and

42Below, I show that there is no NBR for the firms if the search rule does not satisfy theRPP.

52

therefore they don’t search (µj = 0). As in Proposition 1, there is no pure

strategy equilibrium for the firms and the price distribution does not present any

mass points or gaps on the support [p∗c , v]. A firm that sets a price v sells to 1−λ2

captive consumers while if it charges lower prices, the likelihood of capturing the

informed consumers increases. In equilibrium, the firm is indifferent between the

reservation price and lower prices that generate the same expected profit:

πj(pj, c) =

[λ (1− Fc (pj, v)) +

(1− λ)

2

](pj − c) =

(1− λ)

2(v − c) = πj (v, c)

(1.35)

with c ∈ cL, cH. Solving for F (·) , we obtain the following lemma:

Lemma 4 Given the number of shoppers and consumers search rule σD from

(1.34), there is a unique Nash Best Response σF :

Fc(p, v) = 1−(

(1− λ)(v − p)

2λ(p− c)

)p ∈ [p∗c , v] , c ∈ cL, cH (1.36)

with p∗c = c+ (1−λ)(ev−c)1+λ

.

On the demand side, consumers choose their optimal search strategies by

comparing the gains from search (1.33) given σF and production cost prior α.

Using (1.36) and Bayes’ rule, the posterior probability of high cost becomes

α′ =

0 p0 < p∗HαF ′

H

αF ′H+(1−α)F ′

Lp0 ≥ p∗H

(1.37)

Thus, when prices are very low (p0 < p∗H) consumers know that the cost

realization was low and EG (p0 < p∗H , v) increases with prices observed. When the

53

price observed is supported by both, low and high cost of production (p0 ≥ p∗H),

α′ possesses the following properties:

α′ (p0 ≥ p∗H) =α (υ − cH) (p0 − cL)2

α (υ − cH) (p0 − cL)2 + (1− α) (υ − cL) (p0 − cH)2 > α

α′ (p∗H) = α+α (1− α) (cH − cL)

υ − (αcL + (1− α) cH)

∂α′ (p0 ≥ p∗H)

∂p0

< 0

The updated probability jumps from zero to its highest value when p0 = p∗H

and decreases as the observed price is higher. Higher prices are relatively more

likely to be observed when the cost of production is low than when the cost is

high. That is, except at p0 = p∗H , the learning effect reinforces the direct effect

and EG (p0 > p∗H , v) is also monotonic in p0.

To characterize the market equilibrium, the optimal reserve price has to be

consistent with the firms’ NBR. Uninformed consumers are indifferent between

searching or not when the observed price is equal to the reservation price. At

the same time, firms anticipate v correctly and set prices accordingly. Thus, an

equilibrium is characterized by

EG(p0 = v, v) ≤ s (1.38)

Figure 1.10 shows an example with a unique equilibrium.43 Even though EG

is not monotonic in p0, the assumption of small cH − cL relative to s guarantees

a unique reservation price. Intuitively, the closer cH is to cL, the closer p∗H is

43The parameters used are υ = 2, α = 0.5, s = 0.15, cL = 0, cH = 0.5, and λ = 0.2.

54

Figure 1.10: Updated prior and unique reservation price

to p∗L and the array of possible observed prices revealing low cost realization are

never high enough to motivate searching. From the point of view of the firms,

the relevant reservation price is any root in (1.38) that is below υ and over cH .

The next proposition characterizes the market equilibrium.

Proposition 5 Under sequential search and assuming RPP (cH − cL sufficiently

small), there is a unique market equilibrium given by (1.34), (1.36) and (1.37)

and v = Min arg solve(1.38), υ

Proof. Assume wlg cL = 0 and any reservation price v > p∗H . From (1.33) and

Lemma 4, the upper bound for EG (p0 < p∗H , v) is given by

Limp0→p∗H

EG (p0 < p∗H , v) =1

2λ

2λcH + v (1− λ)Log

[v (1− λ)

v (1− λ) + 2cHλ

]This limit increases with cH . Assuming cH sufficiently low, EG (p0 < p∗H , v) < s

and thus v > p∗H . For high observed prices (p0 > p∗H), using p0 = v in (1.33),

EG (v, v) =v (v − cH)

2λ [v − cH (1− α)]

[2λ+ (1− λ)Log

[1− λ

1 + λ

]](1.39)

The expression in brackets is positive for any value of λ and there are two roots

that solve EG (v, v) = s . The smaller one can be ignored since it implies a reser-

vation price below cH .

55

If the assumption on low cH − cL is relaxed, the possibility of non-monotonic

search rules arises (RPP is violated). This can be easily seen in Figure 1.10. As

the gap between the high and low production costs increases, the gap between

the lower bounds for the prices under each cost also increases and the gains from

search could cut the search cost twice. Consider then the case where (1− λ)

uninformed consumers have a search rule q consisting of two reservation prices

(vL, vH):

q =

0 p0 ≤ vL < p

1 vL < p0 < p

0 p ≤ p0 ≤ vH

1 vH ≤ p0

with p > cH . A firm chooses prices given (λ, q) and -when cost realization is

low- takes into account that the (1− λ) /2 potentially captive consumers search

if p ∈ (vL, p) are chosen.

Proposition 6 There is no (symmetric) market equilibrium when consumers

have non-monotonic search rules with two reservation prices.

Proof. By Lemma 3 above, p ≤ vH . A firm setting a price p should ex-

pect profits to be greater than charging the highest reservation price: π (p) ≥

(vH − c) (1− λ) /2. As in Proposition 1, an equilibrium can not consist on pure

strategies (Single Price Equilibrium) nor the price distribution have mass points

or gaps over [p∗, vL] ∪ (p, vH ]. Also, F (vH − ε) < 1 as well as F(p∗cH

+ ε)> 0.

Assume first that F (p′) = F (vL) for p′ ∈ (vL, p) . The expected profits are

π (vH) = (vH − c) (1− λ) /2 (1.40)

π (p) = (p− c) (1− λ) /2 + λ [1− F (p)] p ∈ [p∗, vL] ∪ (p, vH ](1.41)

π (p) = (p− c) (1− λ) /2 + λ [1− F (p)] + F (p)λ/2 (1.42)

56

and there is no F (p) that satisfies (1.40), (1.41) and (1.42) . If p′ ∈ (vL, p) are

considered, expected profit is

π (p′) = (p′ − c)

(1− λ

2+ λ

)[1− F (p′) +

1− λ

2[F (p)− F (p′)]

]Clearly, Lim

p′→epF (p′) < F (p) and thus an atom of probability is needed at p (con-

tradiction).

Summarizing, given the cost gap assumption, consumers’ search strategies

satisfy the RPP and the firms find optimal to set low prices rather than charging

higher prices and observe all the consumers shop around.44 The search intensity in

the market is given by λ and is not affected by changes in the cost of production or

consumers’ prior α. That would lead us to discard any possibility for the rockets

and feathers result. However, consumers play a critical role when shaping the

NBR since their reservation value is a function of the expected prices in the

market hence on α. An increase in α doesn’t affect the search gains when low

prices are observed EG (p0 < p∗H , v) but decreases EG (v, v). Assuming cL = 0,

∂EG(v, v)

∂α=

−cHv − cH (1− α)

EG (v, v) < 0

Therefore the reservation price that equates (1.38) is higher when a higher pro-

duction cost is expected: ∂ev∂α

> 0. This could be interpreted as lower search

intensity in the market since high reservation prices allow the firms to charge

higher prices.

The rockets and feathers result found in Section 1.3 depends on i the stochas-

tic process for the production cost, ii the response on the search intensity (reser-

vation price now) to higher expected costs, and iii the change on the expected

44This is also a consequence of restricting the search cost heterogeneity.

57

cost pass-through by the firms when the search intensity in the market increases.

The equivalent to equation (1.17) would be:

E

[4p4c+

]− E

[4p4c−

]=−1

2ρ (1− ρ)

∂2pt

∂ct∂v

∂v

∂α

∂α

∂ct−1

(1.43)

I now turn to the analysis of the change in the expected cost pass-through as

the reservation price increases. Using 1.36, the expected price is

E [p|c] = υ −eυ∫

p∗(eυ)

F (p, c)dp (1.44)

and for a given reservation price, the jump in prices as cost increases is:

E [p|cH ]− E [p|cL] = (cH − cL)

(1− (1− λ)

2λlog

[1 + λ

1− λ

])> 0 (1.45)

The cost pass-through is not sensitive to consumer’s reservation price and

that means that there is no room for rockets and feathers when nonshoppers are

homogeneous and search sequentially. The following proposition summarizes the

result

Proposition 7 There is no price asymmetry when consumers search sequentially

and have search cost s ∈ 0, s.

The result in the Proposition above is not as negative as it seems. It implies

that a necessary condition for the rockets and feathers under sequential search is

heterogeneity in the search cost of the nonshoppers. Intuitively, in equilibrium,

the nonshoppers with the lower search cost will search (so mu > lambda) and the

58

search intensity changes with the expected production cost. As in the nonsequen-

tial case, the cost pass-through by the firms is sensitive to the search intensity

in the market (see equations 1.45 and 1.19). Therefore, extending the sequential

search model to more general distributions of search costs across consumers is

likely to generate the result.45

45Another possibility (as Jim Dana suggested to me) is to assume a search protocol whereconsumers observe the first price for free and then decide whether to search for k more prices.This protocol is a combination of Dana (1994) and Burdet and Judd (1983) protocols, andmore than two firms in the market are required.

59

CHAPTER 2

Price Dispersion and consumer search.

The Retail Gasoline Markets.

2.1 Introduction

Price dispersion (as opposed to the law of one price) characterizes most final

good markets. With few exceptions, the traditional empirical approach to the

study of this phenomena assumes that product heterogeneity is responsible for

the observed price differences. More recently, findings of significant price disper-

sion in almost homogeneous goods markets expose the need for other potential

explanations and the old theoretical models of costly consumer search were res-

cued by empirical economists.1 There is no doubt that the assumptions of these

models (consumers’s imperfect information and positive search cost) are quite

appropriate to most markets, but researchers have been always skeptic about

the unappealing equilibrium property of firms using mixed-strategies to choose

prices. This paper vindicates the importance of consumers’ imperfect information

to explain the price dispersion observed in retail gasoline markets.

Theoretically, both models of product differentiation and costly consumer

search predict price dispersion in equilibrium. However, the dynamic implications

1Figure 2.4 in the Appendix shows the growth in the number of articles on information,search and price dispersion published in main journals (AER, JPE and Econometrica) duringthe period 1960-2005.

60

of each model are different. In search models, a firm with a high price today might

have the lowest prices tomorrow while in models of product differentiation, this

price dispersion is stable as long as the characteristics of the products don’t

change over time.

The retail gasoline market presents a unique opportunity to identify whether

search is important to explain the observed price dispersion. Gasoline prices are

posted outside each station and many street corners are populated with more

than one of them. The price differences (if any) between gas stations located on

a single corner (control group) can only be related to product differentiation since

consumers are obviously informed about their prices. Moreover, this dispersion

is expected to be stable over time since stations’ characteristics don’t change in

the short run. On the other hand, the price dispersion between stations that are

further apart -but still in the same market- can be generated by both product

differentiation and costly consumer search. If that is the case, we can expect a

more unstable price dispersion for this second group.

This paper provides a simple test of the dynamic stability of price disper-

sion in both groups. The price reversals between gas stations in the same corner

are indistinguishable from zero, as a model of product differentiation would pre-

dict. Conversely, and consistent with an underlying model of costly consumer

search, the price reversals are frequent and important when stations are located

in different corners. Previous empirical work that links price dispersion to costly

consumer search do so by analyzing prices after controlling for product hetero-

geneity. This poses the problem that unobservable characteristics can be behind

the dispersion not explained by observable characteristics. To overcome this po-

tential pitfall, I use a control group of prices not influenced by search.

61

In Chapter 1, I use a theoretical model of consumer search to explain the

asymmetric pricing puzzle (rockets and feathers) observed in many markets. This

asymmetric cost pass-through by the firms is explained by the fact that the

search intensity in a market changes with the past cost realizations. Interestingly,

the rockets and feathers pattern has been found systematically in almost every

study of the retail gasoline market.2 Thus, the findings in this chapter provide

additional support to the theory presented in the previous chapter.

The empirical relevance of costly consumer search has important policy and

academic implications. Outcomes like the rockets and feathers that were thought

to be generated by collusive behavior and requiring government intervention can

well be the consequence of imperfect consumer information. On the academic

side, it emphasizes the need to generalize the traditional structural empirical

models that started with Berry, Levinson and Pakes (1995) to allow for consumer

search. Also, recent work by Klenow and Willis (2006) has shown that traditional

general equilibrium models used to explain price rigidities can not account for

the high frequency of relative prices changes within industries. The main reason

could be found in the fact that such models assume product differentiation only

and ignore the possibility of consumers with imperfect information.

The paper is organized as follows. In the next section, I describe the existing

literature on consumer search and price dispersion. In section 2.3 I describe

the aspects of the gasoline industry that are relevant to this study. Section 2.4

describes the dataset and results. Section 2.5 concludes.

2See Geweke (2004) and Deltas (2004) for surveys of the literature on asymmetric pricingin gasoline markets.

62

2.2 Consumer Search and Price Dispersion

An abundant theoretical literature on price dispersion and consumer search was

developed after Stigler’s seminal paper on economics of information (Stigler,

1961). In that article, Stigler asserts that the price dispersion in the market

is a function of the amount of search conducted by consumers, which at the same

time depends on the nature of the commodity traded. Thus, he argued that

search would be more intense when i) goods represent a larger fraction of the

consumers’ expenditures, ii) the fraction of repetitive consumers is bigger, iii)

the geographic size of the market is smaller (lower search costs). As Rothchild

(1974) pointed out, in Stigler’s model, consumers decided to search taking a cer-

tain distribution of prices in the market as given, and ignored the fact that the

price distribution is endogenous since firms anticipate consumer behavior and set

prices optimally.3 Following Rothschild’s criticism, several economists developed

complete models of costly consumer search that are able to generate price disper-

sion as a stable equilibrium although not necessarily the same comparative statics

Stigler anticipated. In this section, I review the different theoretical approaches

to consumer search and the empirical work that link price dispersion to these

models.

There are many models of consumer search in the literature and their main

results are most of the time driven by the modeling assumptions chosen. However,

the mechanism that generates price dispersion in equilibrium is always the same.

As long as some consumers don’t become completely informed, the firms are

indifferent between high and low prices. By setting high prices, they can sell to

the few consumers with a high search cost (surplus appropriation effect) while

3Another flaw in Stigler’s analysis was ignoring the change in the marginal benefit of a nextsearch by consumers that observe different prices.

63

setting low prices could give them the opportunity to gain market share (business

stealing effect). Table 2.3 in the Appendix provides a summary of the main

differences between the models used in the literature.

One way to explain price dispersion can be by assuming cost dispersion.

The models where firms have heterogeneous cost of production escape from the

Bertrand result where only the most efficient firm serves the market since con-

sumers don’t know the cost realizations and have positive search costs (MacMinn,

1980; Carlson and McAfee, 1983; Benabou and Gertner, 1993). But different

firms’ costs is not necessary to generate dispersion. Burdett and Judd (1983)

were the first to show that the law of one price is not true even when firms and

consumers are completely homogeneous.

Models differ in the method used when obtaining the equilibrium. Stahl

(1996) identifies the Stackelberg and Nash paradigms. Under the Stackelberg

approach, consumers are assumed to know the ”market distribution” of actual

prices being charged but do not know which store is charging which price.4 Under

the Nash paradigm, consumers have less information before search and decide

their strategies based on the Nash Equilibrium pricing strategies of the firms. The

early work on search followed the Stackelberg paradigm (Salop and Stiglitz, 1977;

Braverman, 1980; Rob, 1985; Stiglitz, 1987) but the Nash paradigm as replaced

it as the preferred modeling choice.

Another dimension in which models can be grouped is the search protocol

4For example, if there are N stores whose symmetric mixed-strategy is to draw a randomprice from a probability distribution F (p), these N independent random draws give N actualprices, and the frequency distribution of these actual prices, say M(p), is called the marketdistribution. Note that given a finite number of stores, M(p) can be quite different from F (p).Consumers are assumed to know M(p) and to choose optimal search rules with respect to M(p).

64

used by consumers. A consumer that searches non-sequentially needs to decide

-before observing any prices- between becoming informed about a fixed number of

prices (and buying from the store with the lowest price) or remaining uninformed,

in which case she buys costlessly from a random store. If a consumer were to

search sequentially, after visiting a store she would decide whether to sample

for another price or shop at the lowest price observed at that moment. As it

is well known, both sequential and non-sequential search rules have their own

advantages and disadvantages (Morgan and Manning, 1985). Varian’s model

of sales (Varian, 1980) uses non-sequential search where the sample size is the

number of firms in the market. Others allow for flexible sample size (Burdett and

Judd, 1983) or introduce alternative protocols (Dana, 1994; Janssen, Moraga-

Gonzlez, and Wildenbeest, 2004).5 On the sequential search side, Stahl (1996)

provides the characterization of the equilibrium in a model with heterogeneous

search costs.

Lastly, the distribution of consumers’ search costs has important implications

for the equilibrium properties. For example, the existence of a mass of con-

sumers with zero search cost (also called shoppers) eliminates the possibility of a

monopoly equilibrium or Diamond Paradox (Diamond, 1971): all the firms in the

market charge the monopoly price and not searching is the optimal response by

consumers. Alternatively, models with sequential search where consumers have

the same positive search cost imply no search since firms prefer to decrease the

competition by setting prices below the unique reservation price for consumers

(see Appendix C in Chapter 1).

On the empirical side, the approach taken by most of these studies consists of

5Baye et al. (2004, 2005) show how the non-sequential search model can be thought of as aparticular case of a ”clearinghouse” model of equilibrium price dispersion.

65

two steps. First, they show that after controlling for product heterogeneity, there

is significant price dispersion that needs to be explained. Second, they test some

of the comparative statics (effect of the number of firms and consumer search

costs on the price levels and dispersion) generated by one the theoretical search

models described above, and use that as evidence that the price dispersion in the

first step is indeed generated by consumer search. For example, Sorensen (2000)

finds that the price dispersion and price-cost margins are lower for pharmaceutical

drugs that require repeated purchases than for those that are used infrequently.

Barron et al. (2004) find that increasing the number of sellers decreases price

levels and price dispersion of gasoline prices. Dahlby and West (1986) find also

that the dispersion in insurance premiums is lower for consumers associated with

higher incentives to search. Borenstein and Rose (1994) show that dispersion in

airline fares on a route is higher when more competitors serve the route.6

The problem with most of these studies is that there are always unobserved

characteristics that could explain the corrected or “clean” (from product charac-

teristics) price dispersion. Moreover, the fact that a particular model of search

supports the comparative statics found in the data should not be considered as

additional evidence of search. Almost each theoretical model of search has differ-

ent predictions regarding the response of price dispersion to the number of firms,

search cost level, and other exogenous variables.

The approach taken in this paper is different from previous studies. Due to the

special characteristics of the retail gasoline market, the evidence that search plays

a role in explaining price dispersion is found by comparing the dispersion pattern

generated by a model of product differentiation with one where a priori we are

uncertain about the underlying model. The former group is composed by the

6Baye (2005) provides a comprehensive survey of the empirical work on price dispersion.

66

gasoline stations that are located in a same corner. Clearly, if there is any price

dispersion in this sub market it can not be explained by costly consumer search.

The latter group is composed by all the gasoline stations in a particular market.

We know that product differentiation (at least location) is important to explain

price differences in this market, but are uncertain about search. The key to

identify whether search plays an important role is to compare the stability of the

price dispersion in the two groups. Stable price dispersion is generated by firms

playing pure strategies (product differentiation) but this is not the case when

they use mixed-strategies (costly consumer search).7 Before showing the results,

the next section describes the industry structure that will help to understand the

key assumptions in the econometric analysis.

2.3 The gasoline market

The industry is characterized by refineries that produce gasoline from crude oil

and send it to a main distribution center (distribution rack) to be delivered to its

final destination, the gas station. Gasoline stops being a homogeneous good when

an additive corresponding to the brand of the refiner is mixed with the fuel right

before it is taken for delivery to the gas station. The potential for further differen-

tiation is then increased since gas stations can differ in other dimensions such as

location, capacity, convenience store, car wash, repair facilities, full service, and

methods of payments available to consumers (credit cards, ATM, payment inside

or at the pump). Some studies have measured the importance of these character-

istics and found that the main difference in prices are between branded stations

(Texaco, Shell, BP, Exxon-Mobil) and stations that sell unbranded gasoline or

7As it will be argued in the next section, gas stations have fairly homogeneous costs andtherefore an equilibrium where firms set pure strategies can not be supported by costly con-sumers search.

67

independents (Lewis, 2003; Hastings, 2004).8

Besides different product characteristics, another reason for potential price

dispersion in a similar geographical market is the vertical relationship between

the stations and refiners. Table 2.5 in the Appendix summarizes the industry’s

vertical chain. Branded stations can have three basic vertical contract types with

the branded refiner. The first type is a company operated station (company-op).

The refiner owns the station and an employee of the refiner manages the station.

The refiner sets the retail price directly and pays the employee a salary. The

second type of station is called a lessee dealer. In this case the refiner owns the

station and leases it to a residual claimant. The lessee is responsible for setting

the retail price, but has to purchase wholesale gasoline directly from the refiner

at a price called the Dealer Tank-Wagon price (DTW).9 At the third type of

branded station, a dealer owned station, the retailer owns the station property

and signs a contract with a branded refiner to sell its brand of gasoline. The

retailer can either be supplied directly by the refiner in which case they pay a

DTW, or the dealer can be supplied by an intermediate supplier called ”jobber”.

The jobber purchases gasoline at the distribution rack and pays a wholesale price

called the rack price. That is the refiner’s posted price for branded gasoline at

the distribution rack, and it is the same price for any jobber purchasing at that

rack.10 Independent stations on the other hand can purchase gasoline from any

refiner (branded or unbranded) that sells at the distribution rack.

8Studies on the effect of product differentiation and market power in the gasoline marketalso include Shepard (1991 and 1993), Png and Reitman (1994).

9The DTW is set by the refiner for a particular zone and includes delivery costs. The refineralso sets volume discounts, the lease rate, and other operation stipulations for the station.

10One jobber often supplies, and possibly owns, many different branded and unbrandedstations. Jobbers can purchase branded gasoline and supply it to independent stations if it ischeaper than the unbranded price (the rack prices are “inverted”), but the independent stationcannot post the name of the brand that they are selling.

68

Summarizing, excluding consumers imperfect information, two reasons could

explain differences in prices between two gas stations: product differentiation and

the degree of vertical integration. The last effect implies that firms have different

maximization problems when setting prices and that the wholesale costs (DTW

and branded and unbranded rack prices) could be different across contracts. In

any of these two cases, the price differentials should be constant over time unless

changes in the contracts occur. Shepard (1993) finds that the contract choice

is mainly influenced by the agency costs associated with the characteristics of

different stations. Also, there is evidence that final prices are not affected by

the vertical relations between stations and refiners. Hastings (2004) studied the

acquisition of the independent gasoline retail chain Thrifty by ARCO in Southern

California and finds that the observed increment in prices after the merger was

due to the independents’ lower market share and not a consequence of the change

in contracts (from no integration to partial and full integration).

2.4 Price dispersion analysis

I constructed a unique dataset of retail gasoline prices for more than 2000 stations

in Southern California during the period March 2003 - September 2005. The

dataset was downloaded from public online information where prices are reported

together with the brand, address and city of each gas station.11 Even though the

dataset includes prices for all four grades of gasoline (regular, midgrade and

premium unleaded, and diesel), I concentrate on regular unleaded (87 octanes

grade) since it is the product that accounts for half of the observations. From the

original set of 2367 stations I was able to obtain reliable geographic information

11The source of the information is Oil Price Information Service (OPIS,http//:www.opisnet.com). They collect the price information daily from credit cardtransactions in each store of the sample.

69

Station level

N T Mean sd

Stations∗ 1949 338 days 45 days 34.9

Market coverage 117.017miles2

Mean Min Max sd

Price $2.442 $1.499 $3.499 $0.309

Rank reversal∗∗ 0.1121 0 0.5 0.1574

Spread1= pit − pjt, i 6= j 7.28cts 0 108 5.8918

Spread2=∑

(pit − pjt) /Tij 6.899cts 0 78 6.1895

* Successfully geocoded stations with regular unleaded prices.

** Given a pair of stations (i, j), if most of the time pi ≤ pj then

a rank reversal of x means that pj > pi 100x% of the time.

Table 2.1: Summary Statistics

for 83% of them. Distance is a key element in the analysis that follows so I

discarded any suggested geocoding with a low precision score.12 The panel is

unbalanced in every sense. Not all the stations have prices for the same days nor

have the same number of observations. Table 2.1 summarizes the structure of

the dataset and provides some summary statistics.

A simple measure of price dispersion is the high standard deviation observed

in prices. Also, the average daily price spread between any two gas stations in the

sample is 8cts although it can be as high as one dollar. As explained before price

dispersion can not be used to validate a consumer-search model since gasoline

is also a differentiated product. However, there are differences in the dispersion

12Addresses are not accurate since information is missing or incorrect. I mapped stationsusing a GIS geocoding service and ignored stations that were geocoded with a geocoding scoreof less than 70%.

70

patterns generated under each model. Even though the characteristics of a station

are in the set of choice variables for a firm, the choices over such dimensions

remain fixed for a much longer period of time than prices. If this is the case,

we should expect to observe a fairly stable price dispersion pattern across time

when product differentiation is the only driver of price dispersion. Instead, when

the underlying model involves consumer search, price ranks across stations are

expected to revert as frequently as firms change prices.

The more straightforward way to analyze the stability of price dispersion is

to couple stations and study the behavior of their prices over time. Let sij be

a vector of the price spread between two stations (i, j) over Tij days, such that

pit > pjt is observed most of the time. A measure of instability can be given

by the number of times pjt > pit. The average rank reversals in prices observed

in the dataset is 0.11 (Table 2.1). That means that from the price observations

within a pair of gas stations, the station that usually has the the lowest price

had a high price 11% of the time. By definition, a rank reversal can never be

higher than 0.5. Figure 2.1 shows a histogram of the rank reversals in prices

for all possible pairs of stations that are separated by at most 5 miles from each

other. As it can be seen, for more than 50% of the stations in the sample, the

spread is reverted at least 10% of the time. This is a sign of instability on the

price dispersion pattern that could not be explained by a model where firms have

homogeneous costs and sell differentiated products. But it is consistent with a

model of costly consumer search.

On the other hand, rank changes could be argued to be generated by models

with product differentiation and idiosyncratic costs and demand shocks. If de-

mand shocks are important, firms facing a positive demand shock increase their

prices (and eventually the rank changes) relative to other firms that did not re-

71

02

46

81

0%

of

ob

s

0 .1 .2 .3 .4 .5Rank Reversals

1530 obs | T > 10 | d < 8km

Figure 2.1: Rank reversals in prices

ceive a demand shock. In general, a demand shock is thought of as affecting a

whole market rather than a particular gas station or corner. Thus, if demand

shocks explanained rank reversals we would observe that gas stations in the same

market (less than 1 or 2 miles apart) have lower reversals than those further

apart. In Figure 2.2, the cumulative empirical distributions of rank reversals are

plotted for groups of stations that differ in the distance separating the stations

in each pair. First, the set of stations having at least one competitor within

490 feet were selected. Then, for each distance bound or market area, all pairs

involving one of those stations were formed. It can be seen that the pattern of

rank reversals don’t differ too much when the distance bound is 1, 2 or 5 miles,

but are notably different when stations are separated by at most 490 feet.13 The

price dispersion is more stable between stations that are very close to each other

than those that, while being in the same market, are more distant.14

The remaining potential explanation (besides consumer search) for unstable

13Industry expert as well as most of the empirical papers that deal with retailing gasolineagree in considering the market for one gasoline station to be the area within 1 mile from thestation (Hastings, 2002)

14To avoid the possibility of localized demand shocks like basketball games and other events,only weekday prices were analyzed.

72

.2.4

.6.8

1%

of

ob

s

0 .1 .2 .3 .4 .5Rank Reversals

d < 5 miles (1479 obs) d < 490 feet (51 obs)d < 2 miles (364 obs) d < 1 mile (94 obs)

Figure 2.2: Cumulative rank reversals and distance

price dispersion is the existence of idiosyncratic cost shocks at the station level.

A profit maximizer gas station owner would always use the opportunity cost

of gasoline as the relevant cost for setting prices. As explained in the previous

section, the differences in wholesale costs for two stations are irrelevant in terms of

final prices. But even when they exist, they do for stations that are in different

markets (“zone” prices and delivery costs) and are stable over time since the

contracts with refiners do not change frequently. So wholesale costs are expected

to be highly (if not perfectly) correlated across stations. And that correlation is

supposed to be bigger across gas stations that carry the same brand. But given

that stations with the same brand are never located one across each other, the

correlation in the cost shocks (if any) should increase with the distance separating

the stations. Thus, the group of nearby stations is expected to present more rank

changes than the group of stations that are separated by more than a block. This

is the opposite of what Figure 2.1 shows.

Now assume that a model of costly consumer search and product differentia-

tion together are a good description of the retail gasoline market. Then, a con-

sumer that decided to buy gasoline in station i at price pi is either uninformed or

informed. If she is informed, it means that -after accounting for product differ-

73

entiation spreads- there is no better deal in the route she’s been traveling than

pi. If she is uninformed, station i was picked randomly from the set of stations

(presumably many) she drives by. But, when stations i and k are in front of each

other, consumers are obviously informed of their prices and differences between

pi and pk can only reflect product differentiation.15

We can think of stations i and k as coordinating their prices to compete for

the informed consumers with other distant stations and then compete between

each other for the captive customers based on product differences. In other words,

rank reversals in prices are expected to happen less frequently for stations that

are close from each other than for stations that are nearby but further apart.16

The equality of the observed frequencies of rank reversals in Figure 2.1 can be

tested using a Kolmogorov-Smirnov test. This non parametric test rejects the null

hypothesis of samples coming from the same populations if there exists a point

for which the cumulative empirical distribution of two independent samples are

significantly different. Table 2.2 presents the results. D represents the maximum

distance separating the cumulated empirical distribution of rank changes (rc) for

stations located close to each other (Fc (rc)) with the distribution for stations

within 1 or 2 miles (F1 (rc) and F2 (rc) respectively). In both cases, the null

hypothesis of equal distributions can be rejected at the 0.01 level and lower rank

reversals are observed in the group of clustered stations (Fc > F1 and Fc > F2).

There is no statistical difference between the frequencies of rank reversals between

the groups of stations separated by less than 1, 2 or 5 miles. This is consistent

15See Png and Reitman (1994) for evidence of product differentiation across stations withsimilar location.

16Not all stations that are about 400 feet apart are visible to consumers. At the same time,there is some measurement error in the mapping of the stations and setting a radius below 400feet might eliminate stations that are actually facing each other. Thus positive rank reversalscould actually be observed within this group of stations even without idiosyncratic cost shocks.

74

D p-value

H0 : Fc (rc) < F1 (rc) 0.0847 0.626

Fc (rc) > F1 (rc) -0.3387 0.001

Fc (rc) = F1 (rc) 0.3387 0.001

H0 : Fc (rc) < F2 (rc) 0.0570 0.751

Fc (rc) > F2 (rc) -0.2501 0.004

Fc (rc) = F2 (rc) 0.2501 0.008

Notes: c=490 feet.

Table 2.2: Kolmogorov-Smirnov equality of distributions test

with the hypothesis that cost shocks or demand shocks are not driving the ranking

changes.

There are two things to note from this simple test of rank reversals. First, if

product differentiation is important in this market, the study of rank reversals

might underestimate the presence of costly search in the market. Under pure

product differentiation (and no uncorrelated shocks in demand or costs), the

spread between two prices is likely to remain constant over time. The bigger the

differentiation between two stations, the less likely it is that the prices revert even

if the firms play mixed-strategies. Second, product differentiation should imply

zero rank reversals while the control group shows some of that. This could be

explained by measurement error when geocoding addresses. The distance given

by the GIS software for two neighboring stations sometimes can be off by 1 block.

That is the reason why I used 490feet as the distance bound for the control group.

Additional evidence of search can be found by looking at the relationship

between price dispersion in the market and wholesale cost or price levels. Most

theoretical models predict a reduction in price dispersion as the wholesale cost of

75

Price level and dispersionWeekly, March-August 2005

1.6

1.8

2

2.2

2.4

2.6

2.8

3

3/28/2

005

4/4/20

05

4/11/2

005

4/18/2

005

4/25/2

005

5/2/20

05

5/9/20

05

5/16/2

005

5/23/2

005

5/30/2

005

6/6/20

05

6/13/2

005

6/20/2

005

6/27/2

005

7/4/20

05

7/11/2

005

7/18/2

005

7/25/2

005

8/1/20

05

8/8/20

05

8/15/2

005

8/22/2

005

Ret

ail P

rice

($)

0.04

0.05

0.06

0.07

0.08

0.09

0.1

0.11

0.12

0.13

SD

Retail Price SD retail price

Figure 2.3: Price level and dispersion

production increases. The dataset collected does not allow for careful examination

of this correlation but Figure 2.3 suggets that this is the case. The prices used

correspond to a sample of 500 gas stations for which there where observations

on every Wednesday of each week in the period March to August 2005. As the

figure shows, the standard deviation of those prices increases when the average

price (proxy for wholesale cost) decreases.

2.5 Conclusion

The retail gasoline market is characterized by price dispersion. Theoretically,

both product differentiation and costly consumer search could generate disper-

sion. In this paper I try to find the source of the price dispersion by looking at the

different dynamic implications of each model. Using a simple test that exploits

the particular structure of the gasoline sector, I find that costly consumer search

76

is indeed relevant in this market.

The extent to which price dispersion is explained by consumer search models

has important policy implications. Dispersed prices have different effects on wel-

fare when there is product differentiation than when consumer search is costly.

Under product differentiation, more variety (hence higher price dispersion) in

the market is associated with higher welfare. It is not the same when consumer

search is costly: higher price dispersion implies more search by consumers and

the effect on welfare depends on the relative size of the search costs and the

deadweight loss. It is thus evident that more empirical work aiming at detecting

the underlying model of price dispersion in the market is needed.

77

Appendix

Source: Baye et al. (2005). Social Science Citation Index , Keyword search for”Information OR Price Dispersion OR Search.” 2005* data through third quarter

Figure 2.4: Percentage of articles published in the AER, JPE, and Econometrica

on Information, Search or Price Dispersion

78

Cons. Search Firms Search Demand Num.

Info protocol cost cost firms

Axell (1977) Nash SeqWR Co G(s) q(p) ∞

Baye and Morgan (2001) Nash NS H λ, s > 0 v N

Benabou (1992) Nash Seq G(c) G(s) q(p) ∞

Benabou and Gertner (1993) Nash Seq cL, cH s v 2

Braverman (1980) Stkb NS HU λ, G(s) q(p) N∗

Burdett and Judd (1983) Nash NS-Flex H s > 0 v ∞ / N

Butters (1977) Nash NS & Seq H+ s v ∞

Dana (1994) Nash SNS H s v N

Fishman and Rob (1995) Nash Seq cL, cH s1 < s2 q(p) ∞

Jansen et al. (2004) Nash NS* H λ, s > 0 v N

Carlson and McAfee (1983) Stkb Seq G (c) G(s) v N

MacMinn (1980) Nash NS & Seq G(c) s v ∞

Reinganum (1979) Nash Seq G(c) s q(p) ∞

Rob (1985) Stkb SeqWR H G(s) q(p) ∞

Roshental (1980) Nash NS H λ, s > 0 v N

Salop-Stiglitz (1977) Stkb NonSeq HU s1 < s2 v N∗

Stahl (1989) Nash Seq H s1 < s2 q(p) N

Stahl (1996) Nash Seq H λ, G(s) q(p) N

Stiglitz (1987) Appendix A Stkb Seq H G(s) q(p) N

Stiglitz (1987) Appendix B Stkb SeqWR H G(s) q(p) N

Varian (1980) Nash NS H s1 < s2 v N

Stkb: Stackelberg search protocol; NS: Nonsequential search; NS*: NS with search for first price;

Seq: sequential search; SEqWR: Sequential search with replacement; H: Homogeneous constant

unit cost of production; H+: H with different advertising costs; HU : Homogeneous U-shaped avg. cost.

G (c): Asymmetric (constant) unit costs; cL, cH: Low and high (stochastic) unit cost; co: Convex total .

production cost; s: Homogeneous search cost; G (s): Distribution of costs; λ: Mass of shoppers; s1 < s2

Low and High search costs; υ: Unit demand (up to choke price); q (p): Multiunit demand; N : Fixed finite

number of firms; N∗: Finite, long-run free entry equilibrium.

Table 2.3: Models of search

79

Figure 2.5: Industry structure

80

References

Axell, B. (1977): “Search Market Equilibrium,” The Scandinavian Journal ofEconomics, 79(1), 20–40.

Bacon, R. W. (1991): “Rockets and feathers: the asymmetric speed of adjust-ment of UK retail gasoline prices to cost changes,” Energy Economics, 13(3),211–218.

Barron, J. M., B. A. Taylor, and J. R. Umbeck (2004): “Number of sell-ers, average prices, and price dispersion,” International Journal of IndustrialOrganization, 22(8-9), 1041–1066.

Baye, M. R., and J. Morgan (2001): “Information Gatekeepers on the Inter-net and the Competitiveness of Homogeneous Product Markets,” The Ameri-can Economic Review, 91(3), 454–474.

Baye, M. R., J. Morgan, and P. Scholten (2005): “Information, Search,and Price Dispersion,” in Handbook of Economics and Information Systems.Elsevier Press.

Benabou, R. (1992): “Inflation and Efficiency in Search Markets,” The Reviewof Economic Studies, 59(2), 299–329.

Benabou, R., and R. Gertner (1993): “Search with Learning from Prices:Does Increased Inflationary Uncertainty Lead to Higher Markups,” The Reviewof Economic Studies, 60(1), 69–93.

Berry, S., J. Levinsohn, and A. Pakes (1995): “Automobile Prices in Mar-ket Equilibrium,” Econometrica, 63(4), 841–890.

Borenstein, S., A. C. Cameron, and R. Gilbert (1997): “Do GasolinePrices Respond Asymmetrically to Crude Oil Price Changes?,” The QuarterlyJournal of Economics, 112(1), 305–339.

Borenstein, S., and N. L. Rose (1994): “Competition and Price Dispersionin the U.S. Airline Industry,” The Journal of Political Economy, 102(4), 653–683.

Boyd, M. S., and B. W. Brorsen. (1988): “Price Asymmetry in the U.S.Pork Marketing Channel,” N.C.J. Agricultural Economics, 10, 103–110.

Braverman, A. (1980): “Consumer Search and Alternative Market Equilibria,”The Review of Economic Studies, 47(3), 487–502.

81

Burdett, K., and K. L. Judd (1983): “Equilibrium Price Dispersion,” Econo-metrica: Journal of the Econometric Society, 51(4), 955–970.

Butters, G. R. (1977): “Equilibrium Distributions of Sales and AdvertisingPrices,” The Review of Economic Studies, 44(3), 465–491.

Carlson, J. A., and R. P. McAfee (1983): “Discrete Equilibrium PriceDispersion,” The Journal of Political Economy, 91(3), 480–493.

Dahlby, B., and D. S. West (1986): “Price Dispersion in an AutomobileInsurance Market,” The Journal of Political Economy, 94(2), 418–438.

Dana, James D., J. (1994): “Learning in an Equilibrium Search Model,” In-ternational Economic Review, 35(3), 745–771.

Deltas, G. (2004): “Retail Gasoline Price Dynamics and Local Market Power,”Discussion paper, Working Paper.

Diamond, P. A. (1971): “A model of price adjustment,” Journal of EconomicTheory, 3(2), 156–168.

Eckert, A. (2002): “Retail price cycles and response asymmetry,” CanadianJournal of Economics, 35(1), 52–77.

Fishman, A., and R. Rob (1995): “The Durability of Information, MarketEfficiency and the Size of Firms,” International Economic Review, 36(1), 19–36.

Geweke, J. F. (2004): “Issues in the Rockets and Feathers Gasoline PriceLiterature,” Report to federal trade commission, University of Iowa.

Goodwin, B. K., and M. T. Holt (1999): “Price Transmission and Asym-metric Adjustment in the U.S. Beef Sector,” American Journal of AgriculturalEconomics, 81(3), 630–637.

Hannan, T. H., and A. N. Berger (1991): “The Rigidity of Prices: Evidencefrom the Banking Industry,” The American Economic Review, 84(4), 938–945.

Hastings, J. (2002): “Vertical Relationships and Competition in Retail Gaso-line Markets: Empirical Evidence from Contract Changes in Southern Califor-nia,” POWER Working Papers 075, UCEI.

(2004): “Vertical Relationships and Competition in Retail GasolineMarkets: Empirical Evidence from Contract Changes in Southern California,”American Economic Review, 94(1).

82

Janssen, M. C., and J. L. Moraga-Gonzlez (2004): “Strategic Pricing,Consumer Search and the Number of Firms,” The Review of Economic Studies,71(4), 1089–1118(30).

Janssen, M. C., J. L. Moraga-Gonzlez, and M. R. Wildenbeest (2004):“Consumer Search and Oligopolistic Pricing: An Empirical Investigation,” Dis-cussion Paper 04-071/1.

Karrenbrock, J. D. (1991): “The behavior of retail gasoline prices: symmetricor not?,” Federal Reserve Bank of St. Louis Review, (Jul), 19–29.

Klenow, P. J., and J. L. Willis (2006): “Real rigidities and nominal pricechanges,” Discussion Paper RWP 06-03.

Lewis, M. (2003): “Asymmetric Price Adjustment and Consumer Search: AnExamination of the Retail Gasoline Market,” Discussion Paper 0407010, CSEMworking paperCenter for the Study of Energy Markets.

MacMinn, R. D. (1980): “Search and Market Equilibrium,” The Journal ofPolitical Economy, 88(2), 308–327.

Morgan, P., and R. Manning (1985): “Optimal Search,” Econometrica:Journal of the Econometric Society, 53(4), 923–944.

Neumark, D., and S. A. Sharpe (1992): “Market Structure and the Natureof Price Rigidity: Evidence from the Market for Consumer Deposits,” TheQuarterly Journal of Economics, 107(2), 657–80.

Peltzman, S. (2000): “Prices Rise Faster than They Fall,” Journal of PoliticalEconomy, 108(3), 466–502.

Png, I. P. L., and D. Reitman (1994): “Service Time Competition,” TheRand Journal of Economics, 25(4), 619–634.

Reinganum, J. F. (1979): “A Simple Model of Equilibrium Price Dispersion,”The Journal of Political Economy, 87(4), 851–858.

Rob, R. (1985): “Equilibrium Price Distributions,” The Review of EconomicStudies, 52(3), 487–504.

Rosenthal, R. W. (1980): “A Model in which an Increase in the Number ofSellers Leads to a Higher Price,” Econometrica, 48(6), 1575–1580.

Rothschild, M. (1974): “Searching for the Lowest Price When the Distributionof Prices Is Unknown,” The Journal of Political Economy, 82(4), 689–711.

83

Salop, S., and J. Stiglitz (1977): “Bargains and Ripoffs: A Model of Mo-nopolistically Competitive Price Dispersion,” The Review of Economic Studies,44(3), 493–510.

Shepard, A. (1991): “Price Discrimination and Retail Configuration,” Journalof Political Economy, 99(1), 30–53.

(1993): “Contractual Form, Retail Price, and Asset Characteristics inGasoline Retailing,” The Rand Journal of Economics, 24(1), 58–77.

Sorensen, A. T. (2000): “Equilibrium Price Dispersion in Retail Markets forPrescription Drugs,” Journal of Political Economy, 108(4), 833–850.

Stahl, D. O. (1989): “Oligopolistic Pricing with Sequential Consumer Search,”The American Economic Review, 79(4), 700–712.

(1996): “Oligopolistic pricing with heterogeneous consumer search,”International Journal of Industrial Organization, 14(2), 243–268.

Stigler, G. J. (1961): “The Economics of Information,” The Journal of Polit-ical Economy, 69(3), 213–225.

Stiglitz, J. E. (1987): “Competition and the Number of Firms in a Market:Are Duopolies More Competitive than Atomistic Markets?,” The Journal ofPolitical Economy, 95(5), 1041–1061.

Tirole, J. (1988): The Teory of Industrial Organization. MIT Press, CambridgeMA.

Varian, H. R. (1980): “A Model of Sales,” The American Economic Review,70(4), 651–659.

Verlinda, J. (2005): “Price-Response Asymmetry and Spatial Differentiationin Local Retail Gasoline Markets,” Discussion paper, University of CaliforniaIrvine (Job Market Paper).

84